Delia Bentley, Ricardo Mairal Usón, Wataru Nakamura, Robert D. V - The Cambridge Handbook of Role and Reference Grammar (2023, Cambridge University Press) - Libgen - Li

The Cambridge Handbook of Role and Reference Grammar
Role and Reference Grammar (RRG) is a theory of language in which linguistic

structures are accounted for in terms of the interplay of discourse, semantics
and syntax. With contributions from a team of leading scholars, this Handbook
provides a field-defining overview of RRG. Assuming no prior knowledge, it
introduces the framework step-by-step, and includes a pedagogical guide for
instructors. It features in-depth discussions of syntax, morphology, and lexical
semantics, including treatments of lexical and grammatical categories, the
syntax of simple clauses and complex sentences, and how the linking of syntax
with semantics and discourse works in each of these domains. It illustrates
RRG’s contribution to the study of language acquisition, language change and
processing, computational linguistics, and neurolinguistics, and also contains
five grammatical sketches which show how RRG analyses work in practice.
Comprehensive yet accessible, it is essential reading for anyone who is interested
in how grammar interfaces with meaning.
delia bentley is Professor of Romance Linguistics at the University of

Manchester. She is the author of Split Intransitivity in Italian (2006) and a co-
author of Existentials and Locatives in Romance Dialects of Italy (2015).
ricardo mairal usón is Full Professor of Linguistics at Universidad Nacional

de Educación a Distancia. He has co-authored or co-edited a number of books
including Nuevas perspectivas en Gramática Funcional (1999), which was awarded
the Nation Research Price AEDEAN 1999, and Linguistic Universals (with Juana Gil,
Cambridge, 2006).
wataru nakamura is Associate Professor of Applied Linguistics at Tohoku

University. He is the editor of New Perspectives in Role and Reference Grammar
(2011) and the co-editor of Objectivization and Subjectivization: Typology of Voice
Systems (2012).
robert d. van valin, jr. is Professor Emeritus in the Departments of Linguistics

at the University at Buffalo (SUNY) and the Heinrich Heine University Düsseldorf.
He is the co-author of Functional Syntax and Universal Grammar (with William A.
Foley, Cambridge, 1984), the co-author of Syntax: Structure, Meaning and Function
(with Randy J. LaPolla, Cambridge, 1997), and the author of Exploring the Syntax–
Semantics Interface (Cambridge, 2005).
Published online by Cambridge University Press

cambridge handbooks in language and linguistics
Genuinely broad in scope, each handbook in this series provides a complete
state-of-the-field overview of a major sub-discipline within language study and
research. Grouped into broad thematic areas, the chapters in each volume
encompass the most important issues and topics within each subject, offering a
coherent picture of the latest theories and findings. Together, the volumes will
build into an integrated overview of the discipline in its entirety.
Published titles
The Cambridge Handbook of Phonology, edited by Paul de Lacy
The Cambridge Handbook of Linguistic Code-switching, edited by Barbara E. Bullock
and Almeida Jacqueline Toribio
The Cambridge Handbook of Child Language, Second Edition, edited by Edith L. Bavin
and Letitia Naigles
The Cambridge Handbook of Endangered Languages, edited by Peter K. Austin and
Julia Sallabank
The Cambridge Handbook of Sociolinguistics, edited by Rajend Mesthrie
The Cambridge Handbook of Pragmatics, edited by Keith Allan and Kasia M. Jaszczolt
The Cambridge Handbook of Language Policy, edited by Bernard Spolsky
The Cambridge Handbook of Second Language Acquisition, edited by Julia
Herschensohn and Martha Young-Scholten
The Cambridge Handbook of Biolinguistics, edited by Cedric Boeckx and Kleanthes K.
Grohmann
The Cambridge Handbook of Generative Syntax, edited by Marcel den Dikken
The Cambridge Handbook of Communication Disorders, edited by Louise Cummings
The Cambridge Handbook of Stylistics, edited by Peter Stockwell and Sara Whiteley
The Cambridge Handbook of Linguistic Anthropology, edited by N. J. Enfield, Paul
Kockelman and Jack Sidnell
The Cambridge Handbook of English Corpus Linguistics, edited by Douglas Biber and
Randi Reppen
The Cambridge Handbook of Bilingual Processing, edited by John W. Schwieter
The Cambridge Handbook of Learner Corpus Research, edited by Sylviane Granger,
Gaëtanelle Gilquin and Fanny Meunier
The Cambridge Handbook of Linguistic Multicompetence, edited by Li Wei and Vivian
Cook
The Cambridge Handbook of English Historical Linguistics, edited by Merja Kytö and
Päivi Pahta
The Cambridge Handbook of Formal Semantics, edited by Maria Aloni and Paul
Dekker
The Cambridge Handbook of Morphology, edited by Andrew Hippisley and Greg
Stump
The Cambridge Handbook of Historical Syntax, edited by Adam Ledgeway and Ian
Roberts
The Cambridge Handbook of Linguistic Typology, edited by Alexandra Y. Aikhenvald
and R. M. W. Dixon
The Cambridge Handbook of Areal Linguistics, edited by Raymond Hickey
The Cambridge Handbook of Cognitive Linguistics, edited by Barbara Dancygier
The Cambridge Handbook of Japanese Linguistics, edited by Yoko Hasegawa
The Cambridge Handbook of Spanish Linguistics, edited by Kimberly L. Geeslin
The Cambridge Handbook of Bilingualism, edited by Annick De Houwer and Lourdes
Ortega

The Cambridge Handbook of Systemic Functional Linguistics, edited by Geoff
Thompson, Wendy L. Bowcher, Lise Fontaine and David Schönthal
The Cambridge Handbook of African Linguistics, edited by H. Ekkehard Wolff
The Cambridge Handbook of Language Learning, edited by John W. Schwieter and
Alessandro Benati
The Cambridge Handbook of World Englishes, edited by Daniel Schreier, Marianne
Hundt and Edgar W. Schneider
The Cambridge Handbook of Intercultural Communication, edited by Guido Rings and
Sebastian Rasinger
The Cambridge Handbook of Germanic Linguistics, edited by Michael T. Putnam and B.
Richard Page
The Cambridge Handbook of Discourse Studies, edited by Anna De Fina and Alexandra
Georgakopoulou
The Cambridge Handbook of Language Standardization, edited by Wendy Ayres-
Bennett and John Bellamy
The Cambridge Handbook of Korean Linguistics, edited by Sungdai Cho and John
Whitman
The Cambridge Handbook of Phonetics, edited by Rachael-Anne Knight and Jane
Setter
The Cambridge Handbook of Corrective Feedback in Second Language Learning and
Teaching, edited by Hossein Nassaji and Eva Kartchava
The Cambridge Handbook of Experimental Syntax, edited by Grant Goodall
The Cambridge Handbook of Heritage Languages and Linguistics, edited by Silvina
Montrul and Maria Polinsky
The Cambridge Handbook of Arabic Linguistics, edited by Karin Ryding and David
Wilmsen
The Cambridge Handbook of the Philosophy of Language, edited by Piotr Stalmaszczyk
The Cambridge Handbook of Sociopragmatics, edited by Michael Haugh, Dániel Z.
Kádár and Marina Terkourafi
The Cambridge Handbook of Task-Based Language Teaching, edited by Mohammed
Ahmadian and Michael Long
The Cambridge Handbook of Language Contact: Population Movement and Language
Change, Volume 1, edited by Salikoko Mufwene and Anna Maria Escobar
The Cambridge Handbook of Language Contact: Multilingualism in Population Structure,
Volume 2, edited by Salikoko Mufwene and Anna Maria Escobar
The Cambridge Handbook of Romance Linguistics, edited by Adam Ledgeway and
Martin Maiden
The Cambridge Handbook of Translation, edited by Kirsten Malmkjær
The Cambridge Handbook of Chinese Linguistics, edited by Chu-Ren Huang, Yen-Hwei
Lin, I-Hsuan Chen and Yu-Yin Hsu
The Cambridge Handbook of Intercultural Pragmatics, edited by Istvan Kecskes
The Cambridge Handbook of Role and Reference Grammar, edited by Delia Bentley,
Ricardo Mairal Usón, Wataru Nakamura and Robert D. Van Valin, Jr.

The Cambridge Handbook
of Role and Reference
Grammar
Edited by
Delia Bentley
University of Manchester
Ricardo Mairal Usón
Universidad Nacional de Educación a Distancia
Wataru Nakamura
Tohoku University
Robert D. Van Valin, Jr.
University at Buffalo, The State University of New York
and Heinrich-Heine-Universität Düsseldorf

Shaftesbury Road, Cambridge CB2 8EA, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi – 110025, India
103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467
Cambridge University Press is part of Cambridge University Press & Assessment,

a department of the University of Cambridge.
We share the University’s mission to contribute to society through the pursuit of
education, learning and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781107130456
DOI: 10.1017/9781316418086
© Cambridge University Press & Assessment 2023
This publication is in copyright. Subject to statutory exception and to the provisions
of relevant collective licensing agreements, no reproduction of any part may take
place without the written permission of Cambridge University Press & Assessment.
First published 2023
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Bentley, Delia, editor. | Mairal Usón, Ricardo, 1965– editor. | Nakamura,
Wataru, editor. | Van Valin, Robert D., Jr., editor.
Title: The Cambridge handbook of role and reference grammar / edited by Delia
Bentley, Ricardo Mairal Usón, Wataru Nakamura, Robert D. Van Valin, Jr.
Description: Cambridge, United Kingdom ; New York, NY : Cambridge University
Press, 2023. | Series: Cambridge handbooks in language and linguistics |
Includes bibliographical references.
Identifiers: LCCN 2022056326 (print) | LCCN 2022056327 (ebook) |
ISBN 9781107130456 (hardback) | ISBN 9781107571440 (paperback) |
ISBN 9781316418086 (epub)
Subjects: LCSH: Role and reference grammar. | LCGFT: Essays.
Classification: LCC P166 .C36 2023 (print) | LCC P166 (ebook) |
DDC 415.01/834–dc23/eng/20230313
LC record available at https://lccn.loc.gov/2022056326
LC ebook record available at https://lccn.loc.gov/2022056327
ISBN 978-1-107-13045-6 Hardback
Cambridge University Press & Assessment has no responsibility for the persistence
or accuracy of URLs for external or third-party internet websites referred to in this
publication and does not guarantee that any content on such websites is, or will
remain, accurate or appropriate.

This volume is dedicated to Charles Fillmore and Michael
Silverstein.

Contents
List of Figures page xi

List of Tables xix
List of Contributors xxii
Pedagogical Guide Robert D. Van Valin, Jr. xxiii
Introduction Delia Bentley 1
Part One Overview

1 Principles of Role and Reference Grammar Robert D. Van Valin, Jr. 17
Part Two Topics in RRG: Simple Sentences

2 Lexical and Grammatical Categories in RRG John M. Peterson 181
3 A Conceptually Oriented Approach to Semantic Composition
in RRG Ricardo Mairal Usón and Pamela Faber 218
4 Semantic Macroroles Rolf Kailuweit 242
5 Grammatical Relations Randy J. LaPolla 269
6 Argument Structure Alternations James K. Watters 292
7 Case Assignment Wataru Nakamura 318
8 Morphology in RRG: The Layered Structure of the Word,
Inflection and Derivation Francisco J. Cortés-Rodríguez 368
9 Adverbs, Mimetics and Ideophones Kiyoko Toratani 403
10 Adposition Assignment and Adpositional Phrase Types
in RRG Sergio Ibáñez Cerda 428
11 The RRG Approach to Information Structure Delia Bentley 456
12 Information Structure and Argument Linking Anja Latrouite
and Robert D. Van Valin, Jr. 488

x CONTENTS
Part Three Topics in RRG: Complex Sentences

13 The Structure and Semantics of Complex Sentences Toshio Ohori 525
14 Linking Syntax and Semantics in Adverbial
(Adjoined) Clauses Lilián Guerrero 557
15 Cleft Sentences and Relative Clauses Luis París 591
16 Extraction Restrictions in Complex Sentences Mitsuaki Shimojo 616
Part Four Applications of RRG

17 Role and Reference Grammar and Diachronic Syntax Ranko
Matasović 647
18 Functional Acquisition Processes in Child Language: Links to Role
and Reference Grammar Richard M. Weist 666
19 Grammatical Aspects of Language Processing in the Brain: A Role
and Reference Grammar Perspective Robert D. Van Valin, Jr. 693
20 Formalization of RRG Syntax Laura Kallmeyer and Rainer Osswald 737
21 Computational Implementation and Applications of Role and
Reference Grammar Brian Nolan 785
Part Five Grammatical Sketches

22 A Grammatical Sketch of Cheyenne (Plains
Algonquian, USA) Avelino Corral Esteban 823
23 A Grammatical Sketch of Yimas (Lower Sepik, Papua New
Guinea) William A. Foley 861
24 A Grammatical Sketch of Avatime (Kwa, Niger-
Congo, Ghana) Saskia van Putten and Rebecca Defina 889
25 A Grammatical Sketch of Amele (Papuan, Papua New
Guinea) John R. Roberts 914
26 Case and Voice in Amis (Austronesian, Taiwan) Joy J. Wu 954
Index 983

Figures
1.1 The monostratal nature of RRG page 19

1.2 The layered structure of the clause (preliminary) 23
1.3 The structure of (3) 23
1.4 PrCS and PrDP in English 24
1.5 Constituents of the LSC 27
1.6 The constituent and operator projections 34
1.7 IF indicators in (16) 35
1.8 Examples of the ordering of aspect and tense markers
in different languages 36
1.9 Passive be in (25a) and progressive be and passive be in (25b) 40
1.10 Default placement of phrasal adjuncts in Japanese 43
1.11 Periphery for individual adverb in (37a) 45
1.12 German predicative PPs 49
1.13 Non-predicative PPs in English and Bulgarian 50
1.14 The layered structure of the MP 52
1.15 The structure of the RP in (12a) 53
1.16 Structure of RP with multiple modifiers 54
1.17 General structure of RPs 58
1.18 English RPs 59
1.19 The structure of (10b) from Tagalog 59
1.20 Some syntactic templates for English 60
1.21 Combining syntactic templates 62
1.22 Layered structure of the word following Everett (2002)
and English refusals 63
1.23 The structure of wičháwakat’iŋkte ‘I will beat them to
death’ in (55a, a0 ) 65
1.24 The structure of (55a) 66
1.25 The constituent projection for the multi-juncture
utterance in (63) 70
1.26 The structure of (64e): English clausal coordination 76
1.27 The structure of (65b): English clausal cosubordination
with two shared operators 77

xii LIST OF FIGURES
1.28 The structure of (65c): English clausal cosubordination

with one shared operator 78
1.29 The structure of (60b): Mandarin nuclear cosubordination 78
1.30 Proposed structure for (80b) in Wari’ (Everett 2008) 81
1.31 Interclausal Syntactic Relations Hierarchy 84
1.32 The Interclausal Relations Hierarchy 88
1.33 The constituent, operator and information structure
projections of the clause 115
1.34 Syntactic templates required by the output of Step 3 121
1.35 The output of the semantics-to-syntax linking for (109) 123
1.36 Summary of the linking from semantics to syntax
for (109) 124
1.37 Constructional schema for German ‘plain’ passive 126
1.38 Constructional schema for Sama antipassive 128
1.39 Constructional schema for Japanese plain passive 131
1.40 Constructional schema for Japanese adversative passive 131
1.41 Constructional schema for basic clause structure
in English 134
1.42 Constructional schema for English displaced
wh-questions 135
1.43 Constructional schema for English peripheral adjuncts 136
1.44 Parser output in syntax-to-semantics linking 137
1.45 Output of step 1 in the syntax-to-semantics linking 137
1.46 Output of step 2 in the syntax-to-semantics linking 138
1.47 Output of steps 3 and 4 in the syntax-to-semantics linking 139
1.48 The linking from syntax to semantics in terms of the
algorithm in (128) 140
1.49 The structure of (148b) 154
1.50 Core cosubordination in (142a) 155
1.51 Core coordination in (142b) 156
1.52 Core cosubordination in (143a) 157
1.53 Core coordination in (143b) 157
1.54 Core coordination in (142d) 159
2.1 Constituent projections for (2a), (2b) and (2d) (adapted
from Van Valin 2008: 166) 185
2.2 Structure of RP with MP containing sentential nucleus
(adapted from Van Valin 2008: 173) 188
2.3 Non-predicative (a) and predicative (b) prepositional
phrases in English (Van Valin 2008: 171) 189
2.4 The Kharia lexicon (Peterson 2011a: 78) 191
2.5 TAM/Person- and Case-syntagmas with the same
semantic base 195
2.6 The constituent projection of (47) 197
2.7 Constituent projection of (52) (adapted from Everett
2008: 398) 200
2.8 The sentence base and sentence modality 201
2.9 Constituent and operator projections for two English
sentences (adapted from Van Valin 2005: 14) 203

List of Figures xiii
2.10 Constituent and operator projections in a Japanese

sentence (adapted from Van Valin 2005: 14) 203
2.11 Constituent and operator projections in Kharia 207
2.12 Layered structure of the RP in English (adapted from Van
Valin 2005: 25) 209
2.13 The structure of the Dyirbal sentence in (65b) (adapted
3.1 Tripartite concept lattice (Pustejovsky 2001) 223
3.2 Qualia structure of violin (Pustejovsky and Batiukova
2019: 162) 224
3.3 FunGramKB Ontological Hierarchy for $STERILIZE_00 230
3.4 FunGramKB conceptual entry for PASTEURIZE 232
3.5 FunGramKB Ontological Hierarchy for concepts linked
to $OBSOLETE_00 233
3.6 FunGramKB conceptual entry for $OBSOLETE_00 233
3.7 FunGramKB Ontological Hierarchy for +INVOICE_00 235
3.8 FunGramKB conceptual entry for +INVOICE_00 235
4.1 Actor-Undergoer Hierarchy (modified) 251
4.2 Four types of languages, accusative and ergative
alignment (Haspelmath 2008) 258
7.1 The constituent structure of the simple clause
in English 320
7.2 Actor-Undergoer Hierarchy (adapted from Van Valin and
LaPolla 1997: 146) 322
8.1 The layered structure of the word 372
8.2 The LSW of friends’ 372
8.3 The LSW of the OE complex word bellringestre ‘bell ringer’ 373
8.4 The LSW of a recursive complex word 374
8.5 Feature percolation in RPs 378
8.6 Feature percolation in the LSC 379
8.7 The LSW of Lakhota verb wičhá-wa-k’u 380
8.8 The LSC of Lakhota sentence wičhá-wa-k’u 380
8.9 The LSC of Spanish sentence cantaban villancicos 381
8.10 Exocentric analysis of compound bookseller 392
8.11 Endocentric analysis of compound bookseller 393
8.12 ‘Acentric’ Spanish compound lavaplatos 394
9.1 Example of LSC (modified from Figure 1.13, Van Valin
2005: 22) 405
9.2 Operator projection (adapted from Van Valin 2005: 12) 406
9.3 Structure of (7a) (modified from Figure 1.14, Van Valin
2005: 22) 409
9.4 The mimetic as a nuclear adverb (adapted from Toratani
2007: 333) 420
9.5 The mimetic as a core adverb (modified from Toratani
2007: 334) 421
9.6 Three possible representations for (27c) 422
10.1 Universal oppositions underlying clause structure (Van
Valin 2005: 4) 429

xiv LIST OF FIGURES
10.2 Components of the layered structure of the clause (Van

Valin 2005: 4) 430
10.3 Layered structure of John gave a book to Mary in the library 430
10.4 Actor-Undergoer Hierarchy 432
10.5 English non-predicative PP 435
10.6 English predicative PP 435
11.1 The cognitive states of referents in discourse (Van Valin
and LaPolla 1997: 201) 460
11.2 The Topic Acceptability Scale (Van Valin and LaPolla
1997: 204) 464
11.3 Coding of referents as topic and focus (Van Valin and
LaPolla 1997: 205) 464
11.4 Pragmatically motivated positions in the layered
structure of the clause 465
12.1 Components of the focus structure projection (Van Valin
2005: 77) 489
12.2 (Unmarked) Narrow focus in English 490
12.3 Revised IS-projection 491
12.4 The constituent, operator and information structure
projections of a sentence 491
12.5 A simple example of DRSs 492
12.6 Derivation of predicate focus 493
12.7 Derivation of (unmarked) narrow focus 493
12.8a Representation of pragmatic presupposition for (3b) 496
12.8b Steps 1 and 2 of the semantics-to-syntax linking in (3b) 496
12.8c Step 3 of the semantics-to-syntax linking in (3b) 496
12.8d Step 4 of the semantics-to-syntax linking in (3b) 496
12.8e Summary of linking from semantics to syntax in Japanese
verb-less numeral quantifier construction in (3b) 497
12.9 Summary of linking from syntax and discourse to
semantics in Japanese verb-less numeral
quantifier construction 497
12.10 Linking from DRS in cross-speaker ‘VP’-ellipsis
in English 499
12.11 Constructional schema for English
conjunction reduction 501
12.12 Analysis of conjunction reduction in English 501
12.13 Ungrammatical conjunction reduction in English 502
12.14 Restricted PFD in some SVO languages 512
12.15 Syntactic templates for English wh-question and locative
inversion constructions 512
12.16 Split PFD in Italian 513
12.17 The pervasive role of discourse-pragmatics in grammar 517
13.1 The layered structure of the clause 526
13.2 Clausal coordination in Maori 528
13.3 Core coordination in French 529
13.4 Nuclear coordination in Tukang Besi 530
13.5 Core coordination in Turkish with operator projection 532

List of Figures xv
13.6 Clausal subordination (clause-to-core/complement clause)

in Huallaga Quechua 533
13.7 Clausal subordination (clause-to-periphery/adverbial
clause) in Huallaga Quechua 534
13.8 Clausal subordination (clause-to-noun/relative clause) in
Huallaga Quechua 534
13.9 Core subordination in English 535
13.10 Nuclear subordination in Khwe (Pavey 2010: 235) 535
13.11 Clausal cosubordination in Udihe with
operator projection 537
13.12 Core cosubordination in English 538
13.13 Nuclear cosubordination in French 538
13.14 Sentential coordination in English 540
13.15 Sentential subordination (parenthetical) in English 540
13.16 Sentential subordination (PrDP) in Japanese 542
13.17 Clausal subordination (PrCS) in Japanese 542
13.18 Clausal subordination (clause-to-core periphery/ad-core)
in English 543
13.19 Form–meaning correlation in complex structures 546
14.1 Interclausal Relations Hierarchy (Van Valin 2005: 209) 564
14.2 Conditional ad-clausal subordination in English 571
14.3 Simplified structure of reason relations as sentential
coordination in Yaqui 574
14.4 Sequential temporal relations as ad-core subordination
in Spanish 576
14.5 Sequential temporal relations as core coordination
in Spanish 577
14.6 Simplified structure of sentential subordination
in Spanish 578
14.7 Simplified structure of purposive core cosubordination
in Yaqui 583
14.8 Simplified structure of purposive core coordination
in Yaqui 584
14.9 Nuclear cosubordination in Yaqui 586
14.10 Core cosubordination in Yaqui 586
15.1 Externally headed relative clauses 596
15.2 Internally headed relative clauses (from Van Valin 2012) 598
15.3 Non-restrictive relative clauses 600
15.4 Pronounless relative clauses 602
15.5 Cleft sentence 605
15.6 The contrast between cleft sentences and relative clauses 609
16.1 Potential focus domain in clausal (daughter)
subordination (Van Valin 2005: 214) 619
16.2 Potential focus domain in ad-core subordination (Van
Valin 2005: 216) 620
16.3 Layered structure of the clause and potential
focus domain 623
19.1 Language-related areas of the left hemisphere 696

xvi LIST OF FIGURES
19.2 Simple English transitive linking template 701

19.3 Intransitive template (verbal) 702
19.4 Intransitive template (non-verbal) 702
19.5 Template for three-place predicates 704
19.6 Template for ‘double-object’ construction 704
19.7 Passive template for transitive verbs 705
19.8 Passive template for three-place predicates 705
19.9 English subject wh-Q template 708
19.10 English non-subject wh-Q template 709
19.11 Long-distance wh-question linking template 709
19.12 Infinitival complement template 710
19.13 Croatian linking template for (9) 713
19.14 Croatian linking template for (10) 714
19.15 Bock and Levelt’s model of grammatical encoding
and RRG 724
19.16 Condensed semantics-to-syntax linking 725
19.17 RRG as an abstract model of grammar 728
19.18 RRG as a language processing system 728
20.1 Examples of syntactic representation and syntactic
templates in RRG 738
20.2 Operator marking by features 741
20.3 Periphery marking by features 742
20.4 RRG templates 744
20.5 Elementary RRG trees and their composition 745
20.6 Lexical anchoring of elementary tree templates 745
20.7 Schematic sketch of simple substitution (a) and sister
adjunction (b) 746
20.8 Sister adjunction of periphery elements at
different layers 747
20.9 Two possible syntactic representations of wh-extraction
from complements 749
20.10 Wh-extraction via simple substitution (a) and wrapping
substitution (b) 749
20.11 Wrapping substitution 750
20.12 D-edge in final derived tree for Figure 20.10b 752
20.13 Discontinuous complex predicates 754
20.14 Basic transitive predication template for English
with variants 755
20.15 Example specifications of syntactic fragments 756
20.16 Constituent structure and operator projection for (7) 759
20.19 Encoding the operator structure in node features for (9)
and (10) 761
20.20 Obligatory adjunction of a tense operator 762
20.21 Final feature unifications 763
20.22 Obligatory adjunction of a tense operator: derived tree 764
20.23 Keeping track of the operator projection in edge features 765

List of Figures xvii
20.24 Derived tree for (10) before (top) and after (bottom) final
edge feature unification 766
20.25 RRG template schemas for non-subordinate nexus types 767
20.26 Two ways of compositionally deriving
cosubordination constructions 767
20.27 Core cosubordination via sister adjunction: analysis of (13) 768
20.28 Core cosubordination analysis of (14) derived by
wrapping substitution 769
20.29 Syntactic representations of the examples in (15) 769
20.30 Composition of templates by wrapping substitution for
(15a) and (15b) 770
20.31 Composition of the syntactic representation of (16a) 770
20.32 Subordination via simple and wrapping substitution for
(17a) and (17b), respectively 771
20.33 Derived tree for (18) 772
20.34 Derivation for (18) 773
20.35 Derivation for (19a) 774
20.36 Derivation for (21) 775
20.37 Syntactic tree for (24) 776
20.38 Derivation for (25a) 777
20.39 Linking syntax and semantics: derived tree 778
20.40 Anchoring the default transitive template with
‘smashed’ 779
20.41 Syntactic composition for (26) after argument linking 779
21.1 The Vauquois triangle (based on Dorr et al. 2006: 384) 794
21.2 MT–RRG-based interlingua bridge approach (from Nolan
and Salem 2011: 315) 795
21.3 The RRG interlingua for UniArab (from Nolan and Salem
2011: 315) 795
21.4 The conceptual architecture of the UniArab system (from
Nolan and Salem 2011: 321) 796
21.5 Dialogue management framework for a language
understanding system supporting conversational agents
and a dynamic common ground 800
21.6 Conceptual framework of the conversational software
agent (from Panesar 2017: 190) 801
21.7 The avatar model (from Murtagh 2019b: 95) 802
21.8 The hand articulator in 3D space with nodes on fingers
and thumb (from Murtagh 2019b: 98) 803
21.9 The architecture of FunGramKB (source: www.fungramkb
.com) 806
22.1 Verb structure in Cheyenne (modified from Corral
Esteban 2017: 310) 825
22.2 Layered structure of the clause with constituent and
operator projections 826
22.3 Representation of direct core arguments (cf. 28) 836
22.4 Representation of an applicative construction with a
derived two-place verb 837

xviii LIST OF FIGURES
22.5 Representation of a clause including an argument-

adjunct and an adjunct 837
22.6 Linking from semantics to syntax in Cheyenne 840
22.7 Linking from syntax to semantics in Cheyenne 840
22.8 Information structure-based sentence template 843
22.9 Unmarked predicate-focus structure in Cheyenne 845
22.10 Unmarked sentence-focus structure in Cheyenne 845
22.11 Unmarked narrow-focus structure in Cheyenne 846
22.12 Interaction between the Person and Semantic
Function hierarchies 852
23.1 Clause structure with constituent, operator and
focus projections 866
23.2 Phrase structure with constituent and
23.3 Linking from semantics to syntax in (11) of Yimas 875
23.4 Constituent structure of a non-finite nominalization
in Yimas 878
23.5 Constituent structure of a simple finite nominalization
in Yimas 883
23.6 Constituent structure of a relative clause (28b) in Yimas 886
24.1 The layered structure of the sentence in example (7) 893
24.2 A sentence with pre-detached position and pre-core slot
(example 11) 894
24.3 Sentence from example (20) with constituent and
24.4 Complex RP with constituent and operator projections
(example 24) 900
24.5 Constituent projection of a PP (example 25) 900
24.6 The Interclausal Relations Hierarchy (Van Valin 2005: 209) 907
25.1 Basic syntax of the clause in Amele 916
25.2 The layered structure of the clause 923
25.3 Optional syntactic structures 923
25.4 The basic structure of the RP in Amele 930
25.5 The structure of the possessive RP in Amele 931
25.6 The PrDP and the potential focus domain 933
25.7 The PoDP and the potential focus domain 934
25.8 Semantics-to-syntax linking in a transitive sentence with
acc marking 936
25.9 Semantics-to-syntax linking in an impersonal
verb construction 936
25.10 Semantics-to-syntax linking in a ditransitive predicate 937
25.11 SVC with cosubordinate core juncture 939
25.12 SVC with nuclear subordination 940
25.13 Extended coordinate SVC 940
25.14 Cosubordinate clause chain structure, example (50) 943
25.15 Core subordinate DUn clause, example (56a) 945
25.16 Layered cosubordination in (57) 947

Tables
1.1 Operators page 33

1.2 Operators in the RP 57
1.3 Nexus types 76
1.4 Juncture and nexus in complex RPs 92
1.5 Lexical representations for Aktionsart classes 100
1.6 Valence, macrorole number and M-transitivity 113
2.1 Possible alignments between lexical and syntactic
categories (Himmelmann 2008: 264) 184
2.2 Operators in the layered structure of the clause, from Van
Valin (2005: 9) 202
2.3 Predicate types with respect to ‘finiteness’ marking (Maas
2004: 379, modified) 205
2.4 Operators in the layered structure of the RP (adapted
3.1 Inventory of conceptual logical structures 229
4.1 Logical structures of states and activities and thematic
relations (see Van Valin and LaPolla 1997: 115) 245
4.2 Actor-Undergoer Hierarchy (see Van Valin and LaPolla
1997: 127, 146; Van Valin 2005: 61) 246
4.3 Features and values for activity clusters (from
Kailuweit 2013) 259
4.4 Activity Hierarchy of arguments of selected predicates 261
7.1 Jakobson’s (1936/1984) featural definitions of nominative,
accusative, dative and instrumental cases 335
7.2a Case syncretism in Kabardian 342
7.2b Case syncretism in Palauan 342
7.3 Case syncretism in Romanian 343
7.4 Case marking of O arguments in Estonian 344
7.5 Dative domain in Northern Sámi, Estonian, Finnish and
Tolmači Karelian 351
7.6 Syntax–morphology mapping in Tagalog 354
8.1 Operators in the LSW (a partial classification) 375

xx LIST OF TABLES
10.1 Semantic units underlying the syntactic units of the

layered structure of the clause (Van Valin 2005: 5) 431
10.2 Lexical representations for Aktionsart classes 431
12.1 Interaction of focus structure markedness with
linking markedness 515
14.1 Constructional schema for Yaqui ‘topic chains’ 570
14.2 Constructional schema for ‘I bought frybread in order for
me to eat’ in (27a) 584
18.1 Layered structure of the clause (LSC) and operator scope
in RRG 669
18.2 Lexical representations for four basic Aktionsart
classes 670
18.3 The percentage of verb tokens produced with -ta and with
-te i- by Sumihare for two predicate categories and three
age periods 672
18.4a The average age of the emergence of tense-aspect forms
for (atelic/telic) predicates (future imperfective) 674
18.4b The average age of the emergence of tense-aspect forms
for (atelic/telic) predicates (‘present’ progressive) 674
18.5 The average age of the emergence of tense-aspect, tense,
and aspect contrasts summing over values of
lexical aspect 675
18.6 The percentage of agreement and tense contrasts for the
three predicate types in Polish and English 675
18.7 Melissa Bowerman’s observations of her daughter
Christy’s temporal utterances 678
18.8 The initial occurrence of the temporal adverbs in three
sets of adverbs 679
18.9 Sentence examples in the acquisition of existential
meaning in Korean 680
18.10 Typology of restricted neutralization of semantic roles
(Van Valin and LaPolla 1997: 269) 681
18.11 Average age and range of initial person agreement
contrasts in English and Polish 682
18.12 The acquisition sequence for canonical and non-
canonical constructions 684
20.1 Operators in the layered structure of the clause (cf. Van
Valin 2005: 9) 757
21.1 The German bracket structure construction (based on
Diedrichsen 2014: 109) 790
23.1 Yimas agreement affixes for local persons 871
23.2 Yimas agreement affixes for non-local persons 871
24.1 Paradigms of PSA prefixes (most noun classes omitted) 895
24.2 Characteristic properties of Avatime SVC subtypes 906
24.3 Semantic functions of SVCs in each subtype 911
25.1 M-intransitive posture verbs 917
26.1 Voice markers and the applicative markers in Amis
(adapted from Wu 2006: 289) 969

List of Tables xxi
26.2 The relationship between the juxtaposed verb in the

BLAQ evaluative construction, SVCs and commentative
CCCs in Squlip Atayal: a comparison (from Yeh 2015: 138) 976
TABLEAUS
7.1 Transitive constructions in Icelandic page 333
7.2 ‘Dative-subject’ constructions in Icelandic 333
7.3 Transitive constructions in Warlpiri 334
7.4 Case syncretism in Kabardian 353
7.5 Case syncretism in Old Persian 353

Contributors
Delia Bentley – University of Manchester

Avelino Corral Esteban – Universidad Autónoma de Madrid
Francisco J. Cortés-Rodríguez – Universidad de la Laguna
Rebecca Defina – University of Melbourne
Pamela Faber – Universidad de Granada
William A. Foley – University of Sydney and Columbia University
Lilián Guerrero – Universidad Nacional Autónoma de México
Sergio Ibáñez Cerda – Universidad Nacional Autónoma de México
Rolf Kailuweit – Heinrich-Heine-Universität Düsseldorf
Laura Kallmeyer – Heinrich-Heine-Universität Düsseldorf
Randy J. LaPolla – Beijing Normal University at Zhuhai, China, and Nanyang
Technological University, Singapore
Anja Latrouite – Heinrich-Heine-Universität Düsseldorf
Ricardo Mairal Usón – Universidad Nacional de Educación a Distancia
Ranko Matasović – University of Zagreb
Wataru Nakamura – Tohoku University
Brian Nolan – Technological University Dublin
Toshio Ohori – Keio University
Rainer Osswald – Heinrich-Heine-Universität Düsseldorf
Luis París – Consejo Nacional de Investigaciones Científicas y Técnicas
John M. Peterson – Christian-Albrechts-Universität zu Kiel
Saskia van Putten – Radboud University, Nijmegen
John R. Roberts – SIL International
Mitsuaki Shimojo – University at Buffalo, The State University of New York
Kiyoko Toratani – York University
Robert D. Van Valin, Jr. – University at Buffalo, The State University of New York,
Heinrich-Heine-Universität Düsseldorf
James K. Watters – SIL International
Richard M. Weist – SUNY Fredonia
Joy J. Wu – National Taiwan Normal University

Pedagogical Guide to The
Cambridge Handbook of
Role and Reference
Grammar
These suggestions are for using the Handbook as a text for beginning and
advanced courses in RRG. Exercises for a beginning course can be found in
Emma Pavey’s The Structure of Language: An Introduction to Grammatical Analysis
(Cambridge, 2010). Exercises for both beginning and advanced courses can
be found in Van Valin and LaPolla’s Syntax: Structure, Meaning and Function
(Cambridge, 1997).
Beginning Advanced
Introduction: §1.1 x x
The structure of simple sentences:
Basic clause structure: §1.2–1.2.1 x x
Lexical and syntactic categories: §1.2.2 x (Ch. 2)
Operators: §1.2.3–1.2.3.2, 1.2.3.4 x (Ch. 2) x (Ch. 2)
Status of copular be: §1.2.3.3 x
Peripheries: §1.2.4 x x
Phrasal adjuncts: §1.2.4.1 x (Ch. 10) x (Ch. 10)
Non-phrasal adjuncts: §1.2.4.2 x (Ch. 9) x (Ch. 9)
Structure of phrases: §1.2.5 x (Ch. 10) x (Ch. 10)
Syntactic templates: §1.2.6 x x
Formalization x (Ch. 20)
Structure of words: §1.2.7.0 x x (Ch. 8)
Lexical integrity hypothesis: §1.2.7.1 x
Head-marking: §1.2.7.2 x
Semantic representation of simple sentences

Lexical representation: §1.4.2.1 x x (Ch. 3)
Semantic roles: §1.4.3 x x (Ch. 4)
Information structure
Basic concepts: §1.5 x (Ch. 11) x (Ch. 11)
Linking syntax and semantics in simple sentences:

Grammatical relations x (Ch. 5) x (Ch. 5)
Case assignment x (Ch. 7) x (Ch. 7)
Linking from semantics to syntax: §1.6.1 x x

xxiv PEDAGOGICAL GUIDE
(cont.)
Beginning Advanced
Linking from syntax to semantics: §1.6.2 x x

Constructional schemas: §1.6.3 x x
Constructional schemas and linking: §1.6.4 x
Information structure and linking x (Ch. 12)
Patterns of argument realization x (Ch. 6) x (Ch. 6)
Semantic motivation and
Cross-linguistic variation: §1.6.5 x x
Structure of complex sentences: §1.3 x (Ch. 13) x (Ch. 13)

Juncture: §1.3.1 x (Ch. 13) x (Ch. 13)
Nexus: §1.3.2 x (Ch. 13) x (Ch. 13)
Issues concerning juncture and nexus: §1.3.3 x
Syntax–semantics interface in complex sentences: §1.3.4

Interclausal syntactic relations
hierarchy: §1.3.4.1 x (Ch. 13) x (Ch. 13)
Interclausal semantic relations
hierarchy: §1.3.4.2 x (Ch. 13) x (Ch. 13, 14)
Representing the semantics of
clause linkage: §1.4.2.2 x (Ch. 13) x (Ch. 13)
Representing constructional
meaning: §1.4.2.3 x
Interclausal Relations Hierarchy: §1.3.4.3 x (Ch. 13) x (Ch. 13, 14)
English causative verbs: §1.3.4.4 x
Complex RPs: §1.3.5 x (Ch. 15) x (Ch. 15)
Linking syntax and semantics in complex sentences: §1.7

Modifications of the linking
algorithm: §1.71 x (Ch. 14, 15) x (Ch. 14, 15)
Argument identity in complex
sentences: §1.7.2 x x
Constraints on extraction x (Ch. 16)
Applying RRG
Historical linguistics x (Ch. 17)
Language acquisition x (Ch. 18)
Neurolinguistics x (Ch. 19)
Computational linguistics x (Ch. 21)
Grammatical sketches x (Ch. 22–26)

Introduction
Delia Bentley
Aims and Scope of the Cambridge Handbook of Role

and Reference Grammar
For Role and Reference Grammar (henceforth RRG), capturing the extent of
variation in the grammars of the world’s languages is as important a goal of
linguistic theory as identifying and explaining the properties that all gram-
mars share. The first work which aimed to provide a rigorous, comprehen-
sive and coherent analysis of the syntax of a number of typologically
dissimilar languages, introducing many of the constructs and principles
which would then be developed into tenets of RRG, was Foley and Van
Valin’s (1984) Functional Syntax and Universal Grammar. Almost four decades
after the publication of that volume, the Cambridge Handbook of Role and
Reference Grammar sets forth to fulfil a more ambitious set of objectives. The
volume is an up-to-date presentation of the framework, assuming very little
familiarity, if any, on the part of the reader, while also introducing the
many developments which RRG has undergone since the publication of the
following manuals and collections: Van Valin and LaPolla (1997), Van Valin
(2005), Pavey (2010) and Mairal Usón et al. (2012) (for further relevant work
see the bibliography available here: https://rrg.caset.buffalo.edu/). In pursu-
ing this objective, the volume is meant to be, quite literally, a handbook,
that is, the most complete current treatment of RRG and the first point of
reference for any researchers and teachers interested in this framework. Van
Valin’s Chapter 1 provides a broad overview of the principles and workings
of the framework, while the chapters in Parts II and III adduce more detailed
discussions of simple and complex sentences, respectively. Part II deals with
the lexicon, lexically motivated alternations, the structure of the word, and
semantic decomposition (Cortés-Rodríguez; Mairal Usón and Faber;
Peterson; Watters); semantic macroroles and grammatical relations
(Kailuweit; LaPolla); important facets of the linking (Latrouite and Van
Valin; Nakamura); adpositional, adverbial and mimetic constructions
https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

2 DELIA BENTLEY
(Ibáñez Cerda; Toratani), and the encoding of information structure

(Bentley; Latrouite and Van Valin). In Part III a discussion of the RRG theory
of predicate and clause linkage (Ohori) is followed by chapters on adverbial
sentences (Guerrero), relative clauses and clefts (París) and extraction restric-
tions in complex sentences (Shimojo). Many of these chapters offer specialist
contributions on important issues in syntax, morphology, lexical semantics,
discourse, and the interfaces between these levels of analysis. Therefore,
they will be of interest to researchers who want to compare different
perspectives on specific topics in the study of language.
Although it remains a principal objective of RRG to provide an adequate
set of tools for the description of the syntax of the world’s languages, RRG
now also aims to offer an explanatory framework for the study of language
acquisition, language change and processing and computational linguistics,
and it has made contributions in the domain of neurolinguistics. The
extensions of RRG to these fields are presented in Part IV: Matasović argues
that the constructs of RRG, which were primarily developed to conduct
synchronic work, also lend themselves to capture several aspects of lan-
guage change. Weist explores the acquisition of key components of the
RRG architecture of grammar within a cognition and communication
approach to language acquisition. Van Valin uses the tools of RRG to explain
the ability of split-brain patients to provide grammaticality judgements
with their isolated right hemisphere, developing a proposal which could
potentially also capture the decoupling of grammaticality judgements and
interpretation in agrammatic aphasics. Finally, Kallmeyer and Osswald
develop an RRG system for the formalization of syntactic and semantic
composition operations, with potential advantages for the computational
implementation of the framework, while Nolan discusses the merits of RRG
in natural language processing. This part of the handbook will give the
reader an opportunity to evaluate the potential of RRG in applied domains,
and to gauge the psychological plausibility of the framework, as compared
with approaches which make very different analytical assumptions.
Over the years, the framework has been adopted by a large number of
fieldworkers and scholars committed to the documentation and investiga-
tion of lesser-known and endangered languages. The concluding chapters, in
Part V, are prime examples of this endeavour, featuring grammatical
sketches of languages spoken in Papua New Guinea (Yimas, Lower Sepik,
described by Foley; Amele, Papuan, treated by Roberts), Africa (Avatime,
Kwa, Niger-Congo, discussed by van Putten and Defina) and North America
(Cheyenne, Algonquian, dealt with by Corral Esteban), as well as a treatment
of voice and case in Amis, Austronesian, by Wu.
The handbook is aimed at a very large readership. Along with those who
wish to adopt the RRG framework in their own research, this includes all
advanced researchers in syntax, morphology and the discourse–semantics–
syntax interface, teachers of syntax at higher education institutions, com-
putational linguists, cognitive neuroscientists, and linguistic fieldworkers.

Introduction 3
It is hoped that these different audiences will find the handbook instructive
and stimulating, and that the descriptions and analyses presented here will
engender constructive scientific debate both within the RRG community
and further afield.
The volume is the result of long-term collaborative efforts of many scholars
worldwide. The editors would like to thank the reviewers of the handbook
proposal, and of the individual chapters, the many chapter authors, and last
but by no means least, the whole community of RRG linguists, who enthu-
siastically and unfailingly supported the project through its many stages and
vicissitudes, helping the editors to see it to its successful completion. At
Cambridge University Press, the editors thank Helen Barton, Isabel Collins
and Stephanie Taylor for their prompt, effective and friendly support.
RRG in Modern Linguistic Theory
An important concern of RRG is to engage in cross-theoretical debate. To give

but few examples, many chapters in Van Valin and LaPolla (1997) include
discussions and mentions of relevant work in different theoretical
approaches, while Van Valin (2001) systematically and explicitly compares
the perspectives of Relational Grammar, Principle and Parameters theories
and Lexical Functional Grammar with that of Role and Reference Grammar
(see also Butler 2005a, b). As noted in Van Valin (2009), this type of evaluative
and contrastive debate, which had previously been central in linguistics, has
become less intense since the late 1970s, although we note that Farrell (2005)
and Dalrymple (forthcoming) (including Bentley and Vincent forthcoming)
offer prime examples of such comparisons. In this section, we therefore
reflect on the place of RRG in modern linguistic theory without aiming at
exhaustivity, but rather selecting some of the themes which are at the very
core of the RRG conception of the architecture of grammar and assessing the
perspective of RRG vis-à-vis that of other approaches. The principal character-
istic of RRG which we hope to highlight in the discussion is that it seeks to
explain the similarities and differences in the syntax of the world’s languages
in terms of the interplay of syntax with lexical-semantic and information-
structural representation, and it is with reference to the interfaces that it
makes its predictions. Therefore, RRG is a prime example of those that
Jackendoff (2002) called parallel architecture theories, that is, linguistic theories
which keep syntax separate from the other levels of analysis while placing the
interplay of these levels at the forefront of linguistic investigation. Alongside
Lexical Functional Grammar (Bresnan 1982, 2001; Börjars et al. 2019) and
Head-Driven Phrase Structure Grammar (Pollard and Sag 1994; Sag and
Wasow 1999), RRG is also one of the monostratal syntactic theories, which
means that it does not rely on movement, or promotions and demotions, and,
instead, it derives any cross-constructional comparisons and generalizations
from facets of the linking of syntax with semantics and discourse.

4 DELIA BENTLEY
Whilst placing due emphasis on relational constructs, RRG has distanced

itself since its very inception from the framework which claimed grammat-
ical relations to be primitives of syntactic theory and universals of human
language, namely Relational Grammar (for Relational Grammar see
Perlmutter 1983; Perlmutter and Rosen 1984; Perlmutter and Joseph 1990;
for relevant discussion, see Van Valin 1977, 1981; Foley and Van Valin 1984:
388–389; Van Valin and LaPolla 1997: 242–285). RRG recognizes that individ-
ual languages may privilege the argument which bears unmarked case or is
placed in a particular position in the clause, treating it as the source of
agreement relations, the antecedent in anaphoric relations, and the control-
ler – or the controllee – in interclausal cross-reference relations. However,
no principle of RRG syntax requires that such coding and behavioural
properties, or any language-specific subsets thereof, should cluster together,
thus defining the subject, the object, etc. of a given language (see Chapters 1
and 5). Grammatical relations are thus not universal in RRG (see Chapter 5
and, for a discussion of comparable views, Farrell 2005: 14–38), but rather
have to be defined with reference to specific constructions, where a
restricted number of semantic relations or pragmatic functions is neutral-
ized for syntactic purposes. It is worth pointing out here that although
Lexical Functional Grammar has a comparable notion of semantically unre-
stricted grammatical functions – subject and objects (Börjars et al. 2019:
332), this framework differentiates between restricted and unrestricted
functions, whereas no grammatical relation is postulated in RRG, if there
is no restriction. The restricted neutralizations of semantic relations which
are relevant to the definition of grammatical relations in RRG are captured
with reference to the generalized semantic relations actor and undergoer,
which are universal. These macroroles are defined on the basis of a hierarchy
of five grammatically salient positions in the semantic representation of
predicates (Van Valin and LaPolla 1997: 146). Actor generalizes across pos-
itions at the high end of the hierarchy, while undergoer generalizes across
positions at the low end (see below and Chapters 3 and 4). The alignment of
the privileged syntactic argument of a construction with the actor, or with
the high end of the said hierarchy, characterizes accusative alignment,
whereas ergative alignment aligns the privileged syntactic argument of a
construction with the undergoer, or the low end of the hierarchy. Passive
and, respectively, antipassive voice flag the marked choice in each align-
ment type (see Chapter 26).
While the construction-specific account of grammatical relations readily
captures split alignment, some languages, which are well represented in the
Indo-European family, tend to select the actor as the privileged syntactic
argument consistently across constructions. The notion of subject ensues
precisely from this consistency, and it is in terms of this consistency that it
can be understood and defined in RRG (Van Valin and LaPolla 1997: 175),
although it has no cross-linguistic validity. The subject in English and other
languages is defined by the restricted neutralization (A(ctor of transitive),

Introduction 5
S (actor or undergoer of intransitive), derived-S (of passive)), which, crucially,

leaves out U(ndergoer of transitive). The RRG account of grammatical rela-
tions as the synchronic neutralization, and the diachronic grammaticaliza-
tion, of semantic relations turns out to be cogent in the analysis of all-focus
constructions, where the controller of V-S agreement cannot align with the
topic and is instead purely selected on semantic grounds (Bentley 2018;
Bentley and Cennamo 2022).
Clause structure has distinctive properties which set RRG apart from most
syntactocentric and parallel architecture theories alike. The framework
aims to do justice to the aspects of phrase structure that are comparable
across languages, while also representing the broad range of variation
observed in the phrase structure of the world’s languages. Since all lan-
guages distinguish structurally between predicating and non-predicating
elements, the clause is composed of the following units or layers: (i) the
Nucleus, which hosts the predicate, (ii) the Core, which includes the Nucleus
and the referential phrases required by the predicate (the core arguments),
and (iii) a Periphery for each layer, containing adjunct modifiers of that
layer (see Chapters 1, 9 and 10). These three semantically defined layers are
the only universal components of the clause.
The building blocks of the clause, and hence the nodes of the syntactic
projection called constituent projection, are not named after the major
phrasal categories NP, AP, VP. This is in part the consequence of the
absence of rewrite rules like S ! NP VP, although the main rationale of
this virtually unique feature of RRG syntax is the empirical observation
that noun phrases can predicate, although they are normally referential,
and in fact in some languages they predicate without the support of a verb,
while, in turn, verbs can be referential expressions (Van Valin 2008:
163–164 and Chapter 2 of this volume). Grammatical relations have no
configurational definition in RRG (compare the notions such as Specifier of
IP in Chomskyan generative theories) and the VP is ruled out as a universal
feature of clause structure, as is also the case with Lexical Functional
Grammar (Börjars et al. 2019: 5–6). To capture discontinuous constituency
and the languages with completely unconstrained word order, there is no
no-crossing condition in the constituent projection. In addition, the bound
pronouns of head-marking languages, as well as the person and number
morphology borne by the verb in null-subject languages, are linked to the
Core node in the constituent projection, thus reflecting their referentiality
and, ultimately, their status as core arguments.
As was briefly mentioned, RRG syntax cannot rely on movement. Rather,
each language has an inventory of syntactic templates, which are drawn
upon in parsing to differentiate passive structures from their active coun-
terparts, wh-questions from their declarative counterparts, etc., while the
correspondences between these pairs of structures are captured at the
interfaces of syntax with semantic representation and information struc-
ture. The syntactic templates in the syntactic inventory of each language

6 DELIA BENTLEY
also reflect the linear ordering requirements of the language and the whole
range of clausal layers that it is endowed with (for the non-universal ones,
which symmetrically occur to the left and the right of the Core, see
Chapter 1). These templates are one of the constructional features of the
framework, and we shall return to other such features below.
The scope of aspect, negation, tense, deontic and epistemic modality, etc.
over different layers of the clause is a language universal for RRG (see
Chapter 9 for similar considerations regarding the scope of adverbs and
ideophones). Indeed, Foley and Van Valin (1984: 208–224) (see also Van Valin
and LaPolla 1997: 46–51; Van Valin 2005: 12) make the strong empirical
claim that the linear order of operators in syntax follows from their scope.
This is the Natural Serialization Principle, which was supported by the results
of a typological survey reported in Bybee (1985). Thus, operators with
narrower scope (e.g. aspect) are predicted always to be closer to the nucleus
than operators with wider scope (e.g. deontic modality). Although similar
claims are made in Functional Syntax (Dik 1978, 1980, 1989), as well as in
Cartographic proposals (Cinque 1999), it is essential to note that the con-
stituent projection can only host predicative and referential units in RRG,
and, thus, it cannot accommodate any functional projections, comparable
to IP, or indeed Mood(. . .)P, Tense(. . .)P, Aspect(. . .)P, etc. Instead, the universal,
and strictly hierarchical, array of operators is represented in the operator
projection (see Chapter 1).
The fact that operators are represented separately from the clausal layers
upon which they have scope has brought to light an interesting, and hith-
erto unexplained, property of language change (Matasović 2008 and
Chapter 17). In accordance with the principle of unidirectionality, which
has been explored from many theoretical perspectives, operators of
narrower scope tend to grammaticalize as operators of broader scope (aspect
> tense; deontic modality > epistemic modality). In RRG terms, this means
that nuclear operators can only develop from nuclear operators, but can
themselves develop into nuclear, core or clause operators. In turn, core
operators can only develop from nuclear or core operators, but can them-
selves develop into core or clause operators. This is what Matasović (2008)
calls the centrifugal direction of grammaticalization. However, the gramma-
ticalization of verbal forms proceeds from lexical units to clitics and affixes,
thus moving in a direction which is, in effect, centripetal, and opposite to
that of the change in semantic scope. Thus, by disentangling operators from
the syntactic units upon which they have scope, work in RRG has uncovered
the opposite directions of the two unidirectional processes which character-
ize grammaticalization. This is an important contribution of the framework
to the study of language change.
The layers of clause structure, and the respective operator scope, play a
key role in the RRG theory of predicate and clause linkage (for which see
Chapters 13, 14, 15, 16). This theory relies on a distinction which, to our
knowledge, has no direct equivalent in other frameworks, namely that

Introduction 7
between nexus and juncture. Nexus is the relationship established between

two clausal layers: RRG makes a trifold distinction between coordination,
co-subordination, and subordination, the last of these being further subdiv-
ided into complement and adverbial subordination, in accordance with
other scholarship. Juncture amounts to the issue of which layers are joined
together by one of the nexus types. All in all, there are nine nexus–juncture
combinations, and operator scope is one of the principal diagnostics that
are available to identify them. Not only has RRG produced important
insights on predicate and clause linkage cross-linguistically, for example
by highlighting differences in the number of nexus–juncture combinations
which are available across languages, but it has also advanced knowledge
on the interplay between the semantic relation between the units in a
construction (causative, aspectual, psych-action, purposive, etc.) and the
degree of syntactic cohesion that is established between them. Building
upon Silverstein (1976) and Givón (1980), RRG has in fact developed an
Interclausal Relations Hierarchy (Van Valin and LaPolla 1997: 481–483; Van
Valin 2005: 209, see also Casti 2012), which juxtaposes an array of semantic
relations with a range of nexus–juncture types, both being listed in decreas-
ing order of cohesion. The mapping between the two sides of the
Interclausal Relations Hierarchy is many to one, and, in fact, the hierarchy
brings to light the possibility and tendency for the more cohesive semantic
linkage types (e.g. causation) to be expressed not only by tight morphosyn-
tactic linkages, but also by looser ones. Importantly, RRG makes the strong
falsifiable prediction that the tightest syntactic linkage realizing a particu-
lar semantic relation in a given language should be higher than or as high
as the tightest syntactic linkage realizing lower semantic relations on the
hierarchy in the same language. In light of the pivotal role played by
predicate and clause linkage in the syntax of natural languages, it is to be
hoped that the work conducted in RRG will inspire further research on this
topic in the future.
Although we cannot go into any of the relevant details (for which we refer
the reader to Chapter 8), we wish to mention here that RRG has a theory of
word structure, which parallels its theory of clausal structure. The three
layers Nucleus, Core and Word are projected onto the constituent projection
of the word, while the operators which have scope over each of the word
layers figure in the operator projection of the word. The RRG approach to
inflectional morphology has been characterized as inferential-realizational,
in the sense of Stump (2001), and an interesting debate which has taken
place in recent years concerns this type of morphology. Whereas some place
inflection in the constituent projection of the word as daughters of the Core
layer (see Everett 2002; Van Valin 2013), Martín Arista (2009: 90) proposes
instead to treat inflection as part of the operator projection.
We dedicate the concluding part of this introduction to the interplay of
syntax with lexical semantics and discourse. RRG has a bidirectional
semantics–syntax and syntax–semantics linking, which attempts to

8 DELIA BENTLEY
reproduce the process of language production and language comprehen-

sion, respectively. The steps of the linking that are concerned with seman-
tics, called ‘the lexical phase’, are argued to be more regular and
comparable cross-linguistically than the other steps of the linking, called
‘the morphosyntactic phase’. The claim is therefore that the more seman-
tically motivated a linguistic phenomenon is, the less cross-linguistic
variation there is, and vice versa (see Section 1.6.5). Discourse constitutes
an independent component of grammar, whose role is pervasive in the
linking. Indeed, the interplay of discourse with syntax and semantics is
claimed to be the site of a great deal of cross-linguistic variation (Van
Valin 2014): different languages may encode the same discourse roles in
different ways: prosodically, syntactically, morphologically and even by
particular lexical choices. There is, therefore, no universal association of
syntactic positions or projections with specific discourse functions (see
notions such as Top(ic)P, Foc(us)P, etc.). Such associations are of course
admitted, but only on a language-specific basis. In fact, the only analytical
assumption that the theory makes about the discourse–syntax interface is
that the outermost positions in the layered structure of the clause, which
are not universal, may host topics and afterthoughts, while foci figure closer
to the core than topics do. From this point of view, the RRG treatment of
information structure differs substantially from that of other syntactic
theories, notably Cartography (Rizzi 1997 and subsequent literature), while
other parallel architecture frameworks have not, in the past, devoted as
much attention to the interaction of syntax with discourse as RRG has (see
Zaenen forthcoming for Lexical Functional Grammar).
As will be explained in Chapter 11, the RRG understanding of informa-
tion structure draws heavily upon Lambrecht’s (1994) distinction between,
on the one hand, the role played by each information unit in pragmatic
presupposition and assertion, and, on the other, the status of the discourse
referent of each information unit in the minds of the discourse partici-
pants, including how this status changes in discourse or text. Discourse in
RRG is also crucially involved in the retrieval of arguments and predicates
that are not overtly expressed in syntax, such as the silent predicates of
Japanese and the null arguments of pro-drop languages which lack rich
morphology, for example Chinese. While ruling out phonologically null
elements in its constituent projection, RRG has a Completeness Constraint,
which requires that all the arguments that are specified in semantic repre-
sentation must be represented in syntax, and vice versa. RRG thus resolves
the challenging case of what in other frameworks are analysed as phonolo-
gically silent elements by allowing arguments and predicates to participate
in direct semantics–discourse and discourse–semantics linking. To this
effect, it adopts the formalism of Discourse Representation Theory (Kamp
and Reyle 1993). We shall not continue this discussion here, but rather we
refer to Chapter 12 and the work of Shimojo (2004, 2008, 2016), among
others, for further discussion.

Introduction 9
By way of conclusion, we consider the contribution of RRG to the discus-

sion which has come to be known as the projectionist–(neo-)constructionist
debate (see Levin and Rappaport Hovav 2005 for an overview). RRG clearly
belongs to the projectionist theories of grammar, although constructions
also play a role in the linking, as will be pointed out below. While fully
embracing the programme launched by Gruber (1965) and Fillmore (1968),
which aimed at deriving the syntactic realization of the arguments from
semantic properties which they have qua arguments of a particular verb,
RRG also began to address the problems faced by the theories of thematic
roles earlier than such problems came to the fore in the scientific debate.
We refer in particular to the difficulty of diagnosing thematic roles, and
the vexed issue of the granularity and the ordering of such roles in
thematic role hierarchies.
Thematic roles were never defined intuitively in RRG. Rather, drawing
upon Jackendoff (1976) and Dowty (1979), Foley and Van Valin (1984: 47–63)
considered the semantic relations of arguments to be derivatives of decom-
posed predicate structures. The positions of arguments on the decomposed
structure of their predicate defined a cline of accessibility to each macro-
role status, actor or undergoer, and the syntactic realization of the argu-
ments was captured in terms of these generalized semantic relations.
These ideas were later developed into Van Valin and LaPolla’s (1997:
90–158) fully-fledged theory of lexical-semantic representation and macro-
role assignment, where labels such as agent and patient are nothing but
mnemonics (ibid., p. 116). In this theory, there are only five universal,
hierarchically arranged, thematic positions that are relevant to macrorole
assignment and, therefore, to the syntactic treatment of the arguments. It
is, therefore, our contention that RRG takes care of the vexed questions
mentioned above, viz. the diagnostic problem and the granularity and
ordering issues. Of course, it does so by making precise analytical choices,
and, in particular, by relying on a system of lexical-semantic decompos-
ition which takes the Vendlerian Aktionsart classes as foundational
(Vendler 1967[1957]; see Van Valin 2005: 42 for the addition of semelfac-
tives; Smith 1997).
While we do not think that the criticisms which have been advanced
against theories of thematic roles and thematic hierarchies could justifiably
be raised against RRG, it is undeniable that the linking proposed by this
approach is less economical than the single projection proposed in purely
syntactic theories of event structure (see, by way of example, Borer 2005a–c;
Ramchand 2008). While valuing Occam’s razor in scientific investigation,
RRG seeks to address the question of how the interaction of syntax with
semantics and pragmatics in different grammatical systems can best be
captured. It is the very pursuit of this goal that leads RRG to adopt the less
economical solution. Suffice it to mention the contrast between different
types of alignment: this speaks in favour of the disentanglement of the event
structure hierarchy from its syntactic realization across languages. Whilst

10 DELIA BENTLEY
accusative alignment privileges the high end of the hierarchy in syntax (see
above), ergative alignment privileges the low one. In split alignment, the
high end provides the privileged syntactic argument of some constructions,
though not others. The event structure hierarchy is one and the same in all
grammars, and, indeed, there is reason to consider it to be the linguistic
correlate of a cognitive universal. However, the study of argument realiza-
tion across languages indicates that both ends can be syntactically
unmarked in different languages or constructions, and this, in our view,
supports the parallel architecture approach. Interestingly, evidence from
neuroscience also corroborates the idea that the lexicon is an independent
module in linguistic competence (see Chapter 19). Therefore, the linking
will remain at the very centre of the RRG conception of grammar, and
indeed work in RRG has over the years sought to refine the system of lexical
decomposition which the event-structure hierarchy is a derivative of (Mairal
Usón and Faber 2002; González-Orta 2002; Van Valin and Mairal Usón 2014;
Bentley 2019, among others).
As for the role of constructions, these are templates which define the
unique syntactic, morphological, semantic and pragmatic features of indi-
vidual constructions in individual languages. These templates – called
‘Constructional Schemas’ – need not include any of the general principles
which are valid in the grammar of a given language, or across languages,
but rather are sets of specific instructions which, combined with the
general linking principles, constitute the grammar of a particular lan-
guage. Constructions thus play a different role in RRG than in
Construction Grammar (Goldberg 1995, 2006), and, crucially, they do not
in any way reduce the role of the linking or of its components, including
the lexicon.
In this section we have introduced key aspects of the formalism
developed by RRG to explore how different languages express linguistically
salient meaning and communicate it in context. We have shed light on
similarities and differences with other frameworks, and we have reflected
on the role of RRG in current debates in linguistic theory. We hope that
these observations will encourage researchers of various theoretical per-
suasions to read further.
References
Bentley, Delia. 2018. Grammaticalization of subject agreement on evidence

from Italo-Romance. Linguistics 56(6): 1245–1301.
Bentley, Delia. 2019. The logical structure of verbs of quantized and non-
quantized change. Paper delivered at the International Conference on Role and
Reference Grammar, University at Buffalo (SUNY), 19–21 August 2019.
Bentley, Delia and Michela Cennamo. 2022. Thematic and lexico-aspectual
constraints on V-S agreement: Evidence from Northern Italo-Romance.

Introduction 11
In A. Ledgeway, N. Vincent and J. C. Smith (eds.), Periphrasis and Inflexion in

Diachrony. A View from Romance, 335–361. Oxford: Oxford University Press.
Bentley, Delia and Nigel Vincent (forthcoming). LFG and Role and Reference
Grammar. In Mary Dalrymple (ed.), The Handbook of Lexical Functional
Grammar (Empirically Oriented Theoretical Morphology and Syntax).
Berlin: Language Science Press. https://langsci-press.org/catalog/book/312.
Borer, Hagit. 2005a. Structuring Sense I: In Name Only: Oxford: Oxford
University Press.
Borer, Hagit. 2005b. Structuring Sense II: The Normal Course of Events. Oxford:
Oxford University Press.
Borer, Hagit. 2005c. Structuring Sense III: Taking Form. Oxford: Oxford
University Press.
Börjars, Kersti, Rachel Nordlinger and Louisa Sadler. 2019. Lexical-Functional
Grammar: An Introduction. Cambridge: Cambridge University Press.
Bresnan, Joan. 1982. Control and complementation. In Joan Bresnan (ed.), The
Mental Representation of Grammatical Relations, 282–390. Cambridge, MA: MIT
Press.
Bresnan, Joan. 2001. Lexical-Functional Syntax. Oxford: Blackwell.
Butler, Christopher S. 2005a. Functional approaches to language. In
Christopher S. Butler, María de los Ángeles Gómez-González and Susana
Doval Suárez (eds.), The Dynamics of Language Use: Functional and Contrastive
Perspectives, 3–17. Amsterdam: John Benjamins.
Butler, Christopher S. 2005b. Functionalist theories of language. In K. Brown,
(ed.), Encyclopaedia of Language and Linguistics (2nd ed.), Vol. 1–14, 696–704.
Oxford: Elsevier.
Bybee, Joan. 1985. Morphology. Amsterdam: John Benjamins.
Casti, Francesco. 2012. Testing the Interclausal Relations Hierarchy: Aspectual and
Modal Periphrases in Modern Sardinian. PhD dissertation, University of
Manchester.
Cinque, Guglielmo. 1999. Adverbs and Functional Syntax: A Cross-Linguistic
Perspective. Oxford: Oxford University Press.
Dalrymple, Mary. Forthcoming. The Handbook of Lexical Functional Grammar.
(Empirically Oriented Theoretical Morphology and Syntax). Berlin:
Language Science Press. https://langsci-press.org/catalog/book/312.
Dik, Simon. 1978. Functional Grammar. Amsterdam: North-Holland.
Dik, Simon. 1980. Studies in Functional Grammar. London: Academic Press.
Dik, Simon. 1989. The Theory of Functional Grammar, Part 1. Dordrecht: Foris.
Dowty, David R. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel.
Everett, Daniel. 2002. Towards an RRG theory of morphology. Lecture
delivered at the International Conference on Role and Reference Grammar,
University of La Rioja.
Farrell, Patrick. 2005. Grammatical Relations. Oxford: Oxford University Press.
Fillmore, Charles J. 1968. The case for case. In Emmon Bach and Robert T.
Harms (eds.), Universals in Linguistic Theory, 1–88. New York: Holt, Rinehart
& Winston.

12 DELIA BENTLEY
Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and
Universal Grammar. Cambridge: Cambridge University Press.
Givón, Talmy. 1980. The binding hierarchy and the typology of comple-
ments. Studies in Language 4: 333–377.
Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to
Argument Structure. Chicago: University of Chicago Press.
Goldberg, Adele E. 2006. Constructions at Work. Oxford: Oxford University
Press.
González-Orta, Marta Maria. 2002. Lexical templates and syntactic variation:
The syntax-semantics interface of the Old English speech verb secgan. In
Mairal Usón and Pérez Quintero (eds.), 281–302.
Gruber, Jeffrey S. (1965) Studies in Lexical Relations. PhD dissertation, MIT,
Cambridge, MA.
Jackendoff, Ray. 1976. Toward an explanatory semantic representation.
Linguistic Inquiry 7(1): 89–150.
Jackendoff, Ray. 2002. The Architecture of the Language Faculty. Cambridge, MA:
MIT Press.
Kamp, Hans and Uwe Reyle. 1993. From Discourse to Logic. Hingham, MA:
Kluwer.
Lambrecht, Knud. 1994. Information Structure and Sentence Form: Topic, Focus,
and the Mental Representation of Discourse Referents. Cambridge: Cambridge
University Press.
Levin, Beth and Malka Rappaport Hovav. 2005. Argument Realization.
Cambridge: Cambridge University Press.
Mairal Usón, Ricardo and Pamela Faber. 2002. Functional Grammar and
lexical templates. In Mairal Usón and Pérez Quintero (eds.), 39–94.
Mairal Usón, Ricardo and María Jesus Pérez Quintero (eds.). 2002. New
Perspectives on Argument Structure in Functional Grammar. Berlin: Mouton de
Gruyter.
Mairal Usón, Ricardo, Lilián Guerrero and Carlos González (eds.). 2012. El
funcionalismo en la teoría lingüística. La Gramática del Papel y la Referencia.
Introducción, avances y aplicaciones. Madrid: Akal.
Martín Arista, Javier. 2009. A typology of morphological constructions. In
Christopher Butler and Javier Martín Arista (eds.), Deconstructing
Constructions, 85–116. Amsterdam: John Benjamins.
Matasović, Ranko. 2008. Patterns of grammaticalization and the layered
structure of the clause. In Rolf Kailuweit et al. (eds.), New Applications of
Role and Reference Grammar, 45–57. Newcastle upon Tyne: Cambridge
Scholars Publishing.
Pavey, Emma L. 2010. The Structure of Language. An Introduction to Grammatical
Analysis. Cambridge: Cambridge University Press.
Perlmutter, David M. (ed.). 1983. Studies in Relational Grammar 1. Chicago:
University of Chicago Press.
Perlmutter, David M. and Brian D. Joseph (eds.). 1990. Studies in Relational
Grammar 3. Chicago: University of Chicago Press.

Introduction 13
Perlmutter, David M. and Carol G. Rosen (eds.). 1984. Studies in Relational

Grammar 2. Chicago: University of Chicago Press.
Pollard, Carl and Ivan Sag. 1994. Head-Driven Phrase Structure Grammar.
Chicago, IL: University of Chicago Press.
Ramchand, Gillian. 2008. Verb Meaning and the Lexicon. A First-Phase Syntax.
Rizzi, Luigi. 1997. The fine structure of the left periphery. In Liliane
Haegeman, (ed.), Elements of Grammar, 281–337. Dordrecht: Kluwer.
Sag, Ivan and Thomas Wasow. 1999. Syntactic Theory: A Formal Introduction.
Stanford: CSLI Publications.
Shimojo, Mitsuaki. 2004. Quantifier float and information processing: A case
study from Japanese. Journal of Pragmatics 36: 375–406.
Shimojo, Mitsuaki. 2008. How missing is the missing verb? The verb-less
numeral quantifier construction in Japanese. In Robert Van Valin (ed.),
Investigations of the Syntax–Semantics–Pragmatics Interface, 285–304. Amsterdam:
John Benjamins.
Shimojo, Mitsuaki. 2016. Saliency in discourse and sentence form: Zero
anaphora and topicalization in Japanese. In M. M. Jocelyne Fernandez-
Vest and Robert Van Valin Jr. (eds.), Information Structure and Spoken
Language in a Cross-Linguistic Perspective, 55–75. Berlin: Mouton de Gruyter.
Silverstein, Michael. 1976. Hierarchy of features and ergativity. In R. M. W.
Dixon (ed.), Grammatical Categories in Australian Languages, 112–171.
Canberra: Australian Institute of Aboriginal Studies.
Smith, Carlotta. 1997. The Parameter of Aspect (2nd ed.). Dordrecht: Reidel.
Stump, G. 2001. Inflectional Morphology: A Theory of Paradigm Structure.
Van Valin, Robert D., Jr. 1977. Ergativity and the universality of subjects.
Communication and Linguistics Studies 13: 689–706.
Van Valin, Robert D., Jr. 1981. Grammatical relations in ergative languages.
Studies in Language 5: 361–394.
Van Valin, Robert D., Jr. 2001. An Introduction to Syntax. Cambridge:
Cambridge University Press.
Van Valin, Robert D., Jr. 2005. The Syntax–Semantics–Pragmatics Interface:
An Introduction to Role and Reference Grammar. Cambridge: Cambridge
University Press.
Van Valin, Robert D., Jr. 2008. RPs and the nature of lexical and syntactic
categories in Role and Reference Grammar. In Robert D. Van Valin, Jr. (ed.).
Investigations of the Syntax–Semantics–Pragmatics Interface, 161–178.
Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 2009. Linguistics past and present: A view from the
Rhine. In Zarina Estrada Fernández, Albert Álvarez González and María
Belén Carpio (eds.), Ser lingüista: un oficio diverso y polifacético. Diez años de una
Maestría en Lingüística, 155–164. Hermosillo, MX: Editorial Unison.
Van Valin, Robert D., Jr. 2013. Head-marking languages and linguistic theory.
In Balthasar Bickel, Lenore A. Grenoble, David A. Peterson and Alan

14 DELIA BENTLEY
Timberlake (eds.), Language Typology and Historical Contingency, 91–124.

Van Valin, Robert D., Jr. 2014. On the place of information structure in a
grammar. In Ana Díaz Galán et al., Comunicación, Cognición, Cibernétic@.
Actas del XXXI Congreso de AESLA, 88–108. www.acsu.buffalo.edu/~rrgpage/
rrg.html.
Van Valin, Robert D., Jr. and Randy LaPolla. 1997. Syntax: Structure, Meaning
and Function. Cambridge: Cambridge University Press.
Van Valin, Robert D., Jr. and Ricardo Mairal Usón. 2014. Interfacing the
lexicon and an ontology in a linking algorithm. In María Ángeles Gómez,
Francisco Ruiz de Mendoza and Francisco Gonzálvez-García (eds.), Theory
and Practice in Functional-Cognitive Space, 205–228. Amsterdam: John
Benjamins.
Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca, NY: Cornell University
Press, 97–121. (Previously published in Vendler, Zeno. 1957. Verbs and
times. Philosophical Review 56: 143–160.)
Zaenen, Annie. Forthcoming. Information structure. In Mary Dalrymple
(ed.), The Handbook of Lexical Functional Grammar (Empirically Oriented
Theoretical Morphology and Syntax). Berlin: Language Science Press.
https://langsci-press.org/catalog/book/312.

Part One
Overview

1
Principles of Role and
Reference Grammar
Abbreviations
We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss

ing-rules.php), with the following additions:
A actor MP modifier phrase

AOR aorist NEUT neuter
ASP aspect NMR non-macrorole
ATV active voice NUC nucleus
AUH Actor-Undergoer Hierarchy PN proper noun
AV actor voice PoCS post-core slot
CL clause, clitic, classifier PoDP post-detached position
CS constructional schema POT potential
ECS extra-core slot PrCS pre-core slot
EVID evidential PrDP pre-detached position
FIN final stage of a process or PRED predicator
action
IF illocutionary force PROC process
GER gerund PSA privileged syntactic
argument
INGR ingressive PSBL possible
INT interrogative RE referring expression
IRH Interclausal Relations Hierarchy RLS realis
LS(s) logical structure(s) RP reference phrase
LSC layered structure of the clause S sentence, subject
LSMP layered structure of the SEML semelfactive
modifier phrase
LSPP layered structure of the U undergoer
adpositional phrase

18 ROBERT D. VAN VALIN, JR.
LSRP layered structure of the UV undergoer voice

reference phrase
LSW layered structure of the word XP categorially unspecified
phrase
MEP macro-event property
1.1 Introduction
A linguistic theory is animated by specific issues, the consequences of

which distinguish it from other theories.* The general issues, such as
‘What is a possible human language?’, ‘How is language acquired?’, are
common to most approaches. The specific questions which stimulated the
development of Role and Reference Grammar (RRG) were, ‘What would a
linguistic theory look like if it were based on the analysis of languages with
diverse structures, such as Lakhota, Tagalog, Dyirbal and Barai (Papua New
Guinea), rather than on the analysis of English and similar languages?’ and
‘How can the interaction of syntax, semantics and pragmatics in different
grammatical systems best be captured and explained?’ The two questions
highlight the profound implications of the analysis of typologically diverse
languages for the formulation of a linguistic theory, and they indicate that
the resulting theory will be one in which semantics and pragmatics play
significant roles. In other words, RRG is a theory of the syntax–semantics–
pragmatics interface. Many of the constructs postulated in RRG are rather
different from those in other theories, and this is precisely because of this
starting point. For example, theories starting from English and other
familiar Indo-European languages assume that endocentric phrase struc-
ture of the kind found in X-bar syntax is an appropriate model for syntactic
structure universally, whereas the facts of some Philippine, Amazonian
and Native American languages call this seriously into question and
require a different conceptualization of phrase structure, as will be
shown below.
In terms of the classification of linguistic theories proposed in Jackendoff
(2002), RRG is a parallel architecture theory in which syntax, semantics and
discourse-pragmatics are represented independently, and in which each
may interact directly with another. Numerous examples of these inter-
actions can be found throughout this volume. Three additional theoretical
assumptions deserve mention and will be elaborated on later in this chap-
ter. The first is that RRG is a monostratal theory, in that it posits a single
* This research was supported in part by a Fellowship from the Max Planck Society and from the German Research
Foundation via CRC 991 ‘The structure of representations in language, cognition and science.’ I would like to thank Delia
Bentley, Jürgen Bohnemeyer, Fabricio Gerardi, Ranko Matasović, Wataru Nakamura, Rainer Osswald, Mitsuaki Shimojo, and
Jan Ullrich for comments on an earlier draft.

Principles of Role and Reference Grammar 19
Figure 1.1 The monostratal nature of RRG
syntactic representation for a sentence; there are no syntactic derivations of

any kind. Rather, there is a direct mapping between the syntactic represen-
tation and the semantic representation by the RRG linking algorithm. This is
represented graphically in Figure 1.1.
‘Discourse-pragmatics’ parallels the linking algorithm, and this signifies
that it plays a role in the linking between syntax and semantics, an inter-
action which varies across languages with significant typological conse-
quences. The second theoretical assumption concerns the nature of this
syntactic representation: it is concrete, not abstract, reflecting the actual
order of lexical items and grammatical morphemes. A crucial consequence
of this concreteness is a ban on phonologically null elements in the syntax;
RRG does not permit empty elements in the syntax (i.e. no traces, unpro-
nounced copies of NPs, null pronouns, null light verbs, etc.) These two
assumptions strongly constrain the theory.
The third assumption concerns the nature of the grammar as a whole.
Since the early 1970s there has been an opposition between constructional
approaches (e.g. the Standard Theory (Chomsky 1965), Relational
Grammar) and principles-and-parameters approaches (government and
binding (GB), the Minimalist Program) which deny the theoretical validity
of the notion of grammatical construction. Starting in the late 1980s, a
variety of neo-constructional approaches appeared under the heading of
‘construction grammar’, all of which reject the principle-and-parameters
approach and treat grammatical constructions as central to grammatical
description and linguistic theory. RRG has been constructional from the
outset and therefore is a kind of construction grammar, albeit one which
differs in many ways from ‘name brand’ Construction Grammar. These
assumptions will be discussed further and their consequences illustrated
in this chapter.
Linguistics is an integral part of cognitive science and, accordingly, theor-
ies have implications for the understanding of language acquisition and
language processing. RRG poses the questions ‘Can language acquisition be
accounted for without recourse to an autonomous Language Acquisition
Device?’ and ‘Can a model of grammar that answers the typological and
theoretical questions posed above provide any insights into the neurocogni-
tive processing of language?’ The last chapter of Van Valin and LaPolla (1997)

summarizes RRG work on first language acquisition up to that point, and

the tentative conclusion is that the acquisition of a variety of core grammat-
ical phenomena can be explained in RRG terms, including some for which it
has been argued that there is no evidence available to the child regarding
them (e.g. subjacency; see Van Valin 1998, 2001, 2002; Weist, this volume,
Chapter 18). In the last few years there has been research on applying RRG to
language processing, both computational and neurolinguistic. Computa-
tional implementation of RRG is a relatively new undertaking, and the
results are intriguing (see Nolan, this volume, Chapter 21, Kallmeyer and
Osswald, this volume, Chapter 20). With regard to sentence processing, one
of the distinctive attributes of RRG is the bidirectionality of the linking
between syntax and semantics, which is represented by the double-headed
arrow in Figure 1.1. The RRG linking algorithm maps from semantics to
syntax and from syntax to semantics. This is an idealization of what a
speaker does (semantics to syntax) and what a hearer does (syntax to seman-
tics). Hence the design of the theory makes it readily amenable to psycho-
and neurolinguistic sentence processing models, as argued in Van Valin
(2006, this volume, Chapter 19).
The presentation will proceed as follows. Section 1.2 will be concerned
with the syntactic representation of simple sentences, followed by the
discussion of the syntactic representation of complex sentences in Section
1.3. Section 1.4 is devoted to the semantic representation of simple and
complex sentences. The RRG approach to information structure is briefly
introduced in Section 1.5. The next two sections present the linking algo-
rithm for simple sentences (Section 1.6) and aspects of linking complex
sentences (Section 1.7).
1.2 The Structure of Simple Sentences
1.2.1 Constituent Structure

There are a number of options available to linguistic theories regarding the
representation of syntactic structure: phrase-structure trees (Chomsky
1957, 1965), relational networks (Relational Grammar), X-bar trees
(Chomsky 1970), functional structures (f-structures, Lexical Functional
Grammar), binary-branching X-bar trees (GB, Minimalism), etc. All of the
phrase-structure approaches, regardless of whether they assume X-bar or
binary branching, maintain that crossing branches in a tree is not allowed.
None of these options are compatible with the assumptions laid out in the
previous section. To begin with, relational networks and f-structures, both
of which are based on grammatical relations, are abstract and do not meet
the concreteness requirement. More significant is the incompatibility of
the various phrase-structure approaches with the constraints that RRG
imposes on syntactic representations. RRG posits a single, relatively

concrete syntactic representation for a sentence; no derivations involving

multiple representations or phonologically null elements are permitted.
For approaches in the Chomskyan tradition, many syntactic phenomena
require multiple, derivationally related representations, for example wh-
displacement, free phrase and word order phenomena, and three-place
predicates (binary-branching models only). Moreover, these phenomena
are normally analysed as crucially involving phonologically null elements,
namely, traditional traces (GB) or unpronounced copies of XPs (Minimal-
ism). Hence none of the approaches mentioned so far could be used in RRG,
because they are incompatible with the fundamental assumptions the
theory makes.
In Foley and Van Valin (1984) the first ideas for an RRG-compatible theory
of clause structure, termed the layered structure of the clause (LSC),
were articulated, and subsequent work (Johnson 1987; Watters 1993; Van
Valin 1993) led to the development of more sophisticated versions of the
LSC. The LSC is semantically based, in that the central and universal
constructs are motivated by semantic distinctions. The theory posits that
clause structure is based on two fundamental oppositions, which are
universal and grounded in the fact that language is a system for communi-
cation between human beings: the opposition between predicating and
non-predicating elements, on the one hand, and that between non-
predicating elements that are licensed by the predicating element and
those that are not. In other words, all languages distinguish predicates,
arguments, and adjuncts. This follows from the nature of language as a
system of human communication: one of the communicative functions of
language is to convey information, and this information is composed of
propositions, which involve reference and predication. Languages there-
fore have elements that are specialized for referring (arguments) and
predicating (predicates), and these are further distinguished from the
optional modifiers (adjuncts) that enrich the propositions being communi-
cated. This can be represented as in (1).
(1) [PROPOSITION [PREDICATE X ][ARGUMENT (Y1), (Y2), . . .][ADJUNCT (Z1), (Z2), . . .]]
These semantic units underlie the central syntactic units of the layered
structure: the notion of predicate motivates the nucleus, which houses
the predicating element (the predicator), the core, which consists of the
nucleus plus the arguments of the predicating element, and the peripher-
ies, which house the adjuncts. This can be represented as in (2) and applied
to an English example in (3).
(2) [CLAUSE [CORE (Y1) [NUCLEUS X ] (Y2), . . .][PERIPHERYcore (Z1)] CORE][PERIPHERYclause (Z2)]
CLAUSE]
(3) [CLAUSE [CORE Mary [NUCLEUS ate ] a cookie ][PERIPHERYcore after lunch] CORE]
[PERIPHERYclause despite her diet.] CLAUSE]

Nucleus, core and periphery are syntactic units which are relatively motiv-
ated by the semantic constructs in (1) but are not purely semantic. While in
the prototypical case, the nucleus and the predicating element (predicator)
are co-extensive, it is possible to have non-predicators in a nucleus. A classic
example of this is noun incorporation, in which one of the arguments of the
predicator occurs in the nucleus along with the predicator itself, for
example Lakhota [čháŋ kiŋ [NUC kaksá]] (wood the chop) ‘chop the wood’ vs.
[[NUC čhaŋkáksa]] ‘chop wood’, ‘do wood-chopping’. In the non-incorporated
form the object argument čháŋ kiŋ ‘the wood’ occurs as an independent
argument, whereas in the incorporated form the noun čháŋ ‘wood’ occurs
without an article and is compounded with the predicator kaksá ‘chop’,
yielding čhaŋkáksa. Multisyllabic words in Lakhota tend strongly to have
stress on the second syllable, as with kaksá ‘chop’, and in the incorporated
form the stress has shifted to the second syllable, which is to be expected
due to the incorporation of čháŋ ‘wood’ (which loses its own accent), which
creates a trisyllabic word.
With respect to the core, in the prototypical case, a semantic argument of
the predicator in the nucleus occurs in it, and an XP in the core is a semantic
argument of the predicator. However, there are cases in which a semantic
argument does not occur in the core, for example the agent in a passive
construction appears in the core-level periphery, or a wh-expression may
occur in a position outside of the core in some languages (e.g. Figure 1.4).
Conversely, there are XPs occurring in the core which are not semantic
arguments of the predicator in the nucleus. Examples include dummy
arguments, like those that occur with meteorological verbs in some lan-
guages (e.g. It is raining, Il pleut (French), Es regnet (German)), and ‘raised’
arguments in ‘raising constructions’ (e.g. Mary seems to like okra, in which
Mary occurs as a syntactic argument of the core headed by seems but is a
semantic argument of like).
Peripheries canonically contain adjuncts which modify a specific layer of
the clause. In (3), for example, the temporal adjunct after lunch modifies the
core Mary ate a cookie and occurs in the core-level periphery, and the conces-
sive adjunct despite her diet modifies the clause Mary ate a cookie after lunch and
occurs in the clause-level periphery. There is at least one case, however,
where the XP in a periphery is not an adjunct modifier but rather a semantic
argument of the predicator in the nucleus, namely, the agent in a passive
construction. Moreover, there are constructions in which the adjunct repre-
sents a participant in the state of affairs which is not a semantic argument
of the nucleus, for example the deputative beneficiary marked by for in Bill
went to the store for Mary.
The distinctions in (2) are better represented in a tree diagram, as in
Figure 1.2.
The nucleus contains the predicator and may be modified by an adverb in
the nuclear-level periphery.1 The core minimally contains the nucleus (and

Figure 1.2 The layered structure of the clause (preliminary)
Figure 1.3 The structure of (3)
its periphery, if present), which would be the case for predicators which
have no arguments (e.g. Lakhota maǧážu ‘it is raining’), but usually there are
one or more argument expressions. It may optionally be modified by adverbs
or adjunct adpositional phrases in the core-level periphery. The clause
contains the core (plus its periphery, if present), and, like the nucleus and
core, can be modified optionally by peripheral adverbs or adpositional
phrases. The sentence node reflects the traditional idea that a sentence
may consist of multiple clauses; its function will be clarified below. Linear
order is not relevant to the definition of the units of the LSC; the order of
elements in Figure 1.2 was chosen for clarity and is in no way privileged.
This figure presents the universal features of the LSC.
The structure of (3) is given in Figure 1.3. Eat is the predicator in the
nucleus, Mary and a cookie are the core arguments, and the two PPs, after
lunch and despite her diet, occur in peripheries. There is no VP in the structure,
despite the well-known evidence that some constructions in English clearly
involve a ‘VP’-like grouping, for example I expected to find someone washing the
dishes, and [‘VP’ washing the dishes] was Bill. This is because ‘VPs’ are not
universal; some languages have constructions with them and others do

Figure 1.4 PrCS and PrDP in English
not, and so from an RRG perspective, ‘VP’ is a derivative category which

occurs in particular constructions and whose existence in a language is the
result of the interaction of the LSC with other components in the grammar.
Specific examples will be discussed later in this chapter, as well as in
Latrouite and Van Valin (this volume, chapter 12), where an analysis of
‘VP’-ellipsis in English is presented. The two PPs modify different layers of
the clause: the temporal PP after lunch tells when the event in the core
occurred,2 while the concessive PP despite her diet modifies the clause con-
taining the core and the core-level periphery (see Section 1.2.4). The order of
the PPs is not free, as the ordering reflects the layering: the higher-level
modifier follows the lower-level modifier.
The constituents in Figure 1.2 are semantically motivated and universal.
There are other constituents which are not semantically motivated and
accordingly are not universal. The English example in Figure 1.4 illustrates
two of them.
It is important to notice that there is no empty element in the core
representing the displaced wh-expression what; there is no trace or unpro-
nounced copy of it. The linking algorithm will link what directly to the
correct place in the semantic representation for its interpretation; see
Section 1.6.
The position that displaced wh-expressions occur in when they appear at
the beginning of a clause, as in English, German, Icelandic and many other
languages, is known as the Pre-Core Slot (PrCS). It is not restricted to wh-
expressions; non-wh XPs functioning as contrastive topic, as in (4a), or
contrastive focus,3 as in (4b), can appear in the PrCS.
(4) a. A: What do you think of the new novels by Suzanne Collins and John
Grisham?
B: Collins’ book I really enjoyed, but Grisham’s I found boring.
b. A: Have you read anything good lately?
B: Suzanne Collins’ new book I really liked, but John Grisham’s latest
offering I didn’t care for.

In English the subject is the first (non-oblique) argument in the core, and
so a non-wh XP appearing immediately before it is outside of the core and is
therefore in the PrCS. XPs in the PrCS, regardless of whether they are wh-
expressions or not, are not set off by an intonation break, and they never co-
occur with a resumptive pronoun in languages like English.4
The existence of the PrCS in languages with nucleus-initial default word
order (e.g. Tagalog (Austronesian, Schachter and Otanes 1972); Tzotzil
(Mayan, Aissen 1987); and Wari’ (Chapakuran, Brazil, D. Everett 2008)) is
straightforward to diagnose: an XP before the nucleus without an inton-
ation break or an independent resumptive pronoun is in the PrCS. This is
illustrated by the Tagalog examples in (5) from Schachter and Otanes (1972:
496–498).
(5) a. T<um>awa siya sa kaniya kahapon.

<av.rls>laugh 3sg.nom dat 3sg.obl yesterday
‘She laughed at him yesterday.’
b. Kahapon siya t<um>awa sa kaniya.
yesterday 3sg.nom <av.rls>laugh dat 3sg.obl
‘It was yesterday she laughed at him.’
b′. Sa kaniya siya t<um>awa kahapon.
dat 3sg.obl 3sg.nom <av.rls>laugh yesterday
‘It was at him she laughed yesterday.’
The default word order in Tagalog is given in (5a), in which the nucleus
containing the predicator tumawa ‘laughed’ occurs initially in the clause.
It is followed by siya ‘3sg.nom’, which is a clitic pronoun that must
occur in second position in the clause; in (5a) second position is imme-
diately following the nucleus. In (5b) the temporal adverb kahapon ‘yes-
terday’ occurs before the nucleus, and it is followed by siya, the second
position clitic. Based on clitic placement and a number of other criteria,
Latrouite and Van Valin (2021) show that the adverb is in the PrCS in
(5b). The same considerations lead to the conclusion that sa kaniya ‘at
him’ in (5b′) is likewise in the PrCS. It should be noted that the PrCS
element has a contrastive narrow focus reading like the English
example in (4b).
Languages with nucleus-final or extremely flexible word order present
something of a challenge with respect to establishing the existence of a
PrCS, because there is no fixed reference point for determining the onset of
the core, unlike in Tagalog or English. The best evidence is the obligatory
occurrence of wh-expressions at the beginning of the clause in wh-questions.
It is well known, however, that nucleus-final languages rarely have obliga-
tory displacement of wh-expressions to the beginning of the clause; they
either have only wh-in situ or optional displacement. The latter situation,
namely optional displacement to the beginning of the clause or in situ, is in
principle distinguishable from ‘free’ word or phrase order, since in the latter
situation a wh-expression could occur in any position in the clause; in

contrast, with optional displacement, it can occur in only two positions, in

situ or clause-initial. This is evidence for a PrCS, albeit weaker than the
evidence in languages with obligatory displacement.
Many nucleus-final languages, and some with other word orders as well,
have a Post-Core Slot (PoCS), which, like the PrCS, is outside the core but
inside the clause. It is multifunctional from a cross-linguistic perspective. In
Japanese, according to Shimojo (1995), focal XPs, like those in the PrCS in
English, cannot occur in the PoCS, but in Dhivehi (Indo-Aryan; Cain and Gair
2000), wh-expressions can appear in situ (6a) or in the PoCS (6a′), and contrast-
ive focal XPs occur in it as well (6b, b′).5
(6) a. Alı̄ kı̄ke bunı̄ ta?

Ali what say.pst.foc q
‘What did Ali say?’
a′. Alı̄ bunı̄ kīke ta?
Ali say.pst.foc what q
‘What did Ali say?’
b. Māle ulunı̄ma aharen bonı̄ AIS KURĪMU.
˙
Male be.pst.prog.when 1sg drink.pres.foc ice cream
‘When in Male, it is ice cream that I eat.’
b′. Māle ulunı̄ma ais kurı̄mu bonı̄ AHAREN.
˙
Male be.pst.prog.when ice cream drink.pres.foc 1st
‘When in Male, it is I who eat ice cream.’
In Figure 1.4, there is an XP (as for Mary) before the wh-expression, and it
has quite different properties from XPs in the PrCS. It is normally set off by
an intonation break, represented orthographically by a comma, and if the
XP is an argument of the predicator in the nucleus of the clause, as in this
sentence, there is obligatorily a resumptive pronoun in the core of the
clause, in this case she. This position is termed the Pre-detached position
(PrDP). It is usually termed ‘the left-detached position’ in the linguistic
literature, including previous work in RRG. However, there is no left or
right in spoken language, only before and after. Only in written language
can one speak of left and right, and the label ‘left-detached position’ reflects
the fact that European languages are written from left to right. For scholars
writing in languages like Arabic and Farsi, on the other hand, it is the ‘right-
detached position’. There is no such problem with PrCS and PoCS, since ‘pre-’
and ‘post-’ refer to order in speech, not order on the printed page. Accord-
ingly, the problematic terms ‘left-detached position’ and ‘right-detached
position’ are replaced by PrDP and Post-detached position (PoDP), in
order to avoid confusion.6
The two detached positions are outside of the clause and are direct
daughters of the sentence node. Both are normally set off by an intonation
break, and both require a resumptive pronoun if the XP in the detached
position is an argument of the predicator in the nucleus of the clause.
Tagalog presents clear evidence that the PrDP is outside of the clause, as
exemplified in (7) from Schachter and Otanes (1972: 489).

(7) Si May (ay), kailan ba (siya) ba~balik dito?

nom.pn (top) when q (3sg.nom) ipfv.return here
‘As for May, when will (she) come here?’
The topic expression, si May, is set off by an intonation break and option-
ally marked by the topic marker ay (Nuhn 2021; Latrouite and Van Valin
2021). It must be in the PrDP, because it is followed by the adjunct wh-
expression kailan ‘when’, which is in turn followed by two second-position
clitics, the interrogative marker ba and the optional resumptive pronoun
siya ‘3sg.nom’. The placement of the clitics shows that kailan ‘when’ must be
clause-initial, namely in the PrCS, as in (5b,b′), and this means that si May is
outside of the clause and therefore in the PrDP. The structure of (7) is given
in (8).
(8) [S [PrDP Si May (ay)], [CL [PrCS kailan ]¼ba¼(siya) [CORE [NUC babalik] dito?] CL] S]
The constituents of the LSC, both obligatory and optional, minus the per-
ipheries, are given in Figure 1.5.
These constituents can be viewed in terms of whether they are universal
or non-universal. It has been argued that the nucleus, core and clause
‘spine’, plus peripheries, is semantically motivated and is based ultimately
on the necessary properties of language as a system of human communi-
cation. All languages make use of the nucleus, core and clause spine. On the
other hand, not all languages have detached positions or special positions
outside of the core and inside the clause. They are pragmatically motivated
rather than semantically motivated. While it would be impossible to form
propositions without predicators and arguments, it is possible to communi-
cate effectively without these special positions. Lakhota, for example, has a
PrDP but no PrCS or PoCS. Dyirbal, with its extremely (grammatically) free
word order, has no fixed positions with reference to which the PrDP or PrCS
could be defined; moreover, it lacks markers indicating topics and foci akin
to Japanese wa and ga, for example, which could be associated with these
distinctions (see Shimojo 2011). The packaging of information in terms of
special fixed positions for topics and foci is a useful but not an indispensable
feature of human language.
Figure 1.5 Constituents of the LSC

1.2.2 Lexical and Syntactic Categories

In Figures 1.2 and 1.5 the abbreviation ‘XP’ is used in the possible core
argument positions, because the syntactic category of the phrase is not fixed;
it could be a noun or determiner phrase, an adpositional phrase, a core or a
clause. In this sense it is a variable over syntactic categories, like in the term
‘X-bar syntax’, and it will continue to be used in this way. In Figures 1.3 and
1.4, on the other hand, the XPs are Mary, a cookie, what and she, and on
standard assumptions these would be analysed as NPs or determiner phrases
(DetPs).7 What this assumes here is the endocentric nature of syntactic cat-
egories: a noun projects an NP, a verb projects a VP, a preposition projects a
PP, etc. Most syntactic theories assume that most, if not all, syntactic categor-
ies are endocentric; for theories which assume Merge as the central operation
in syntax all syntactic categories are necessarily endocentric.
RRG rejects endocentrism as a fundamental feature of constituent struc-
ture. Rather, the two key notions are the nucleus, which, viewed from a
cross-linguistic perspective, is not restricted to any particular lexical
category or even to being a head (it can be phrasal), and reference phrase
(RP), which is a potentially referring expression, which may be headed in
principle by a range of lexical categories (Van Valin 2008). In the examples
examined thus far, the predicator in the nucleus has been a verb, but this is
not the only possibility, as the examples in (9) show.
(9) a. Max is a very good lawyer. Predicator ¼ a very good lawyer [RP]
b. Max is (extremely) tall. Predicator ¼ (extremely) tall [ADJP]
c. Max is in the house Predicator ¼ in the house [PP]
Be is not the predicator in these sentences; it is an auxiliary required for

nucleus formation when the predicator is non-verbal or passive. If it were
claimed that be is the predicator, then it would be necessary to claim that in
languages which lack such a verb there is a phonologically null verb be which
serves as the predicator. This, however, is a violation of the concreteness
restriction in RRG. Consequently, the predicator in (9a) is the RP a very good
lawyer, in (9b) the adjective phrase (extremely) tall, and in (9c) the PP in the house.
These examples illustrate the fact that the nucleus is not a projection of verb,
or any other category, but is category-neutral, and moreover they also show
that the nucleus need not be a head but can be a phrase. A more extreme
example of a phrasal nucleus will be given in §1.3.3.2.
The usual label for argument expressions is NP, or in some approaches,
DetP. The problem with labelling referring expressions as NPs can be seen in
(10) from Tagalog and (11) from Lakhota.
(10) a. [CORE [NUC T<um>awa ] [RP ang bata] CORE] Tagalog

<av.rls>laugh nom child
‘The child laughed.’
b. [CORE [NUC Bata ] [RP ang tumawa] CORE]
child nom laughed
‘The one who laughed is a child.’

(11) a. [CLAUSE [RP Wíŋyaŋ kiŋ hé] [NUC wačhí-kte-šni]CLAUSE]8 Lakhota

woman the that dance-pot-neg
‘That woman will not dance.’
b. [CLAUSE [RP Wačhí kiŋ hé] [NUC wíŋyaŋ-kte-šni ]CLAUSE]
dance the that woman-pot-neg
‘That dancer will not be a woman.’
In (10a) the predicator in the nucleus is a verb, and the single core
argument is a noun, suggesting that it is correctly analysed as an NP. In
(10b), on the other hand, the predicator is a noun and the head of the
argument expression is the verb tumawa ‘laughed’. Importantly, there is no
change in the form of the words in the two sentences: bata ‘child’ is an
argument in (10a) and the predicator in the nucleus in (10b), and tumawa
‘laughed’ is the predicator in the nucleus in (10a) and the head of an
argument expression in (10b). This contrasts with the English translation
of (10b), in which the verb-based argument expression is expressed as a
relative clause and the nucleus requires the auxiliary be due to the non-
verbal predicator. Thus, unlike in English, a verb can head a referring
expression without modification, and a noun can function as the predicator
without modification. The same pattern holds in the Lakhota sentences in
(11); both predicators take the same inflectional morphology, regardless of
the lexical category of the head, and the head of the ‘NP’ wačhí kiŋ hé ‘that
dancer’ is the verb wačhí ‘dance’.
These examples are no problem for the category-neutral notion of
nucleus, and they show that a category-neutral syntactic category for
referring expressions is needed as well. Van Valin (2005: 28) made an initial
suggestion which was further developed into the notion of RP in Van Valin
(2008).9 It might be objected that there is no need for this notion, because
these examples can be handled in terms of the already existing notion of
DetP. The argument expressions in (10) and (11) would be DetPs, not NPs,
and the complement to the determiner head of the phrase could be of
different categories. However, DetP is a problematic construct, given RRG’s
assumptions about syntactic structure. In order to have a DetP there must
be a determiner functioning as the head of the DetP, and in many lan-
guages there are minimal determiner systems; Russian, Mandarin and
Japanese, for example, have demonstratives but no articles. In such lan-
guages many sentences have arguments which lack determiners, and
therefore to justify calling them DetPs it would be necessary to posit
phonologically null determiners as heads of the DetPs. This is incompat-
ible with the RRG prohibition against phonologically null elements in the
syntax. RP, on the other hand, requires no such postulation of null deter-
miners. Thus in (10) and (11) the arguments are RPs, regardless of whether
the head is a noun or a verb.
Modifiers, in particular those modifying within RPs, do not necessarily
follow an endocentric pattern. The following examples are from D. Everett
(2008).

(12) a. The God is dead philosophers are mostly dead.

b. My grandson likes to give me the who’s the boss now, silly old grandpa wink
frequently.
In (12a) the modifier in the RP is a clause, while in (12b) it is a sentence.

Neither can be analysed as a kind of adjective phrase headed by a lexical
adjective, unlike the very tall man, and accordingly, there is a need for a
category-neutral construct for modifiers analogous to the nucleus for predi-
cators. Van Valin (2008) proposed the notion of modifier phrase (MP) to
accommodate attributive modifiers of all types, as well as adverbial modi-
fiers and adpositional phrase modifiers. Thus, the primary syntactic categor-
ies in RRG are RP, nucleus, PP and MP.10
Phenomena like those in (10) and (11) have led some analysts to suggest
that these languages do not differentiate nouns from verbs, and therefore
roots are precategorial and are assigned a category on the basis of how they
are used in a clause. This is problematic, because it confuses category and
function, as well as lexical and syntactic categories. Assuming endocentrism
and associated phrase structure, bata ‘child’ in (10b) heads a VP, a syntactic
category, and because syntactic categories are projections of lexical categor-
ies, bata ‘child’ must be a verb in this sentence. In (10a), on the other hand,
the same reasoning leads to the conclusion that bata is a noun, since it is
heading an NP functioning as an argument. In order to avoid claiming that
there are two lexical items, one a noun and the other a verb, one solution is
simply to avoid assigning lexical categories to words.11
This is very problematic, because there is evidence that both Tagalog and
Lakhota differentiate nouns from verbs, albeit weakly. Himmelmann (2008)
distinguishes nouns from verbs in Tagalog based on a morphological criter-
ion: verbs can take voice morphology, while nouns cannot. Boas and Deloria
(1941: 23) note that there are morphophonological differences between
nouns and verbs in Lakhota, for example vowel contraction (nouns tend to
contract adjacent vowels while verbs do not), reduplication (common with
verbs, very rare with nouns). There are morphosyntactic differences as well.
The examples in (13) are the same as those in (11) except for the plurality of
the core argument.
(13) a. [CLAUSE [RP Wíŋyaŋ kiŋ hená] [NUC wačhí-pi-kte-šni] CLAUSE] Lakhota
woman the those dance-pl-pot-neg
‘Those women will not dance.’
a′. [CLAUSE [RP *Wíŋyaŋ-pi kiŋ hená] [NUC wačhí-pi-kte-šni] CLAUSE]
woman-pl the those dance-pl-pot-neg
‘Those women will not dance.’
b. [CLAUSE [RP Wačhí-pi kiŋ hená] [NUC wíŋyaŋ-pi-kte-šni ] CLAUSE]
dance-pl the those woman-pl-pot-neg
‘Those dancers will not be women.’
b′. [CLAUSE [RP *Wačhí kiŋ hená] [NUC wíŋyaŋ-pi-kte-šni ] CLAUSE]
dance the that woman-pl-pot-neg
‘Those dancers will not be women.’

An interesting feature of Lakhota is the lack of marking number on a noun

when it is an argument; plurality can be signalled by means of a plural
demonstrative in an RP or a quantifier, but it cannot be marked directly on
the noun itself. This is illustrated in (13a, a′): that wíŋyaŋ ‘woman’ is plural is
indicated primarily by the plural demonstrative hená ‘those’ and also by the
plural agreement on the predicator in the nucleus, but marking plural
directly on wíŋyaŋ renders the sentence ungrammatical, as in (13a′). How-
ever, when the RP argument is headed by a verb, as in (13b, b′), plural
marking on it is obligatory, as in (13b). The sentence is ungrammatical
without it, as in (13b′). It is not the case that nouns like wíŋyaŋ ‘woman’
can never take plural marking; when they function as the predicator in the
nucleus of the clause, as in (13b), plural marking is obligatory, just as it is for
verbs with the same function, as in (13a). Thus nouns and verbs are treated
differently with respect to number marking when functioning as arguments
(referring expressions) but not when they serve as predicators. Hence, lan-
guages like Tagalog and Lakhota do distinguish nouns from verbs, albeit
weakly. The issue, rather, is the existence of restrictions on what function a
lexical item can have.
How can RRG approach the issue of lexical categories, given the range of
cross-linguistic variation, from languages like Tagalog and Lakhota, which
seem to make only weak distinctions between nouns and verbs, to languages
like Latin, Russian and Dyirbal, which make strict divisions between the two
main categories, with English somewhere in the middle? One possibility is
to claim that at the most basic level, lexical items fall into one of two classes:
they are either referring expressions (REs) or predicates. This is related
to the fundamental opposition motivating the layered structure of the
clause, namely, the opposition between predicating and non-predicating
elements. This distinction derives from the nature of language as a system
of communication: communication involves conveying information, which
involves propositions, which involve reference and predication, hence REs
and predicates. REs and predicates can have one of three grammatical
functions: argument, predicator or modifier. These functions are
related to the three syntactic categories in RRG, namely, RP, nucleus and
MP. In (10a) and (11a) a predicate is functioning as the predicator in the
nucleus and an RE as the core argument, whereas in (10b) and (11b) an RE is
the predicator in the nucleus and a predicate serves as the core argument.
The traditional categories of verb, adjective, adverb and adposition are
semantically predicates and grammatically predicators or modifiers, as
defaults. The traditional category of noun encompasses REs serving as argu-
ments as their default function.12
Defining lexical categories in terms of their morphophonological and
morphosyntactic properties leads to the observation that Tagalog, Lakhota
and other such languages show that languages need not make ‘deep’ lexical
category distinctions, and this seems to follow from the ability of lexical
items to function as an argument, predicator or modifier rather freely. This

ability is an important morphosyntactic property which many or all lexical

items share, and it is precisely the morphosyntactic properties that different
words do not share that is the basis for assigning them to different
lexical categories.
In languages with well-defined lexical categories there are strict con-
straints on the grammatical function that a given lexical category may have:
nouns (REs) can be arguments, verbs (predicates of certain semantic types)
can be predicators, and adjectives and adverbs (predicates of certain seman-
tic types) can be modifiers. These restrictions are among the morphosyntac-
tic properties which define the category. If a category is to have a different
grammatical function, for instance noun or adjective as predicator or verb
as argument, special morphosyntactic treatment is necessary, such as the
use of an auxiliary verb or verbalizing derivational morphology for non-
verbal predicators, or nominalizing derivational morphology for non-
nominal arguments. These special treatments are also part of the properties
defining the different categories. This pattern is found in many European
languages, and it has been taken as the norm for human language. This view
has led some linguists, when confronted with Tagalog-type languages, to
posit derivational morphology for verbalization and nominalization, all of
which is marked by zero morphemes. Thus, on this type of analysis, in (10b)
bata ‘child’ has undergone zero-marked verbalization, yielding ‘to be a
child’, and tumawa ‘laughed’ has undergone zero-marked nominalization,
yielding ‘the one who laughed’. The RRG analysis of (10b) given above does
not invoke any zero-marked derivational morphology.
Languages also differ as to the semantic domains encompassed by the
nouns and verbs in them. While words referring to objects, entities,
people, animals, etc., will typically be realized as nouns, and those refer-
ring to actions, events, states of affairs, etc., will typically be instantiated
as verbs, others may be realized in different ways in different languages.
Consider property expressions that indicate characteristics of entities and
objects (e.g. tall, little, red, black, rich, happy, etc.). In some languages (e.g.
English) they constitute a word class all their own, namely, adjectives. In
other languages, however, they do not constitute an independent word
class but rather are treated as a subtype of noun or verb. In Dyirbal (Dixon
1972) property words are assimilated to the category of noun and are
treated morphosyntactically like nouns. In Lakhota, on the other hand,
they are construed as a subclass of verbs and take the same inflectional
morphology as verbs, as illustrated in (14).
(14) Wíŋyaŋ kiŋ hená háŋska-pi-kte-šni. Lakhota13

woman the those tall-pl-pot-neg
‘Those women will not be tall.’
The status of lexical categories in RRG is grounded in the universal

semantic distinction between REs and predicates, which underlies the
noun–verb dichotomy. The lexical categories beyond noun and verb in a

language are differentiations of the functions of predicates as modifiers (i.e.

as adjectives and adverbs), and all of the distinctions must be justified
morphophonologically and/or morphosyntactically. Adpositional predicates
can be arguments, as with verbs like put, or adjunct modifiers. Thus, from an
RRG perspective, lexical categories are language-specific but with a universal
semantic foundation. Moreover, they play a rather different role in the non-
endocentric syntax of RRG than they play in the endocentric syntax of
other theories.
1.2.3 Operators
In Figure 1.4, which represents the structure of As for Mary, what did she eat
after lunch?, the auxiliary verb did is not attached to the LSC tree in any way.
The reason for this is that the LSC trees introduced so far contain only
lexical words and morphemes. Closed-class grammatical morphemes indi-
cating functional categories like aspect, tense, modality and mood (illocu-
tionary force) are not represented. These notions are termed operators in
RRG, and they modify specific layers of the LSC. The operators discussed in
RRG are given in Table 1.1.
A detailed discussion of them can be found in Peterson (Chapter 2 of this
volume), and the emphasis in this section will be on selected cases of the
interaction of operators and the LSC.
The most commonly encountered operators cross-linguistically are aspect,
deontic modality, negation, tense, and illocutionary force (IF).14 Following a
suggestion in Johnson (1987), they are represented in a separate projection
of the clause (the operator projection) which is more or less a mirror
image of the lexical tree, which is normally referred to as the constituent
projection. A revised version of the tree in Figure 1.4 with the operator
projection added is given in Figure 1.6.
Table 1.1 Operators
Nuclear operators
Aspect
Negation
Directionals (only those modifying the orientation of an action or event without reference to
participants)
Core operators
Directionals (only those expressing the orientation or motion of one participant with reference to
another participant or to the speaker)
Event quantification
Modality (root modals, e.g. ability, permission, obligation)
Internal (narrow scope) negation
Clausal operators
Status (epistemic modals, external negation)
Tense
Evidentials
Illocutionary force (IF)

Figure 1.6 The constituent and operator projections15
The operator projection is not an exact replica of the constituent projec-

tion in two ways. First, it contains only the nucleus-to-sentence spine and
does not include the other positions and the peripheries. Second, if there is
more than one operator at a given layer, that node is duplicated so that each
operator modifies a distinct node in the operator projection. In this
example, there are two clause-level operators, tense and IF, each modifies a
separate clause node, and the structure makes clear that IF has scope over
tense. The evidence for this will become clear in the discussion of complex
sentences in Section 1.3.
Both operators are linked to the auxiliary did, because it expresses both of
them, tense via its form (past tense of do) and IF by virtue of its position in the
clause. English signals IF by means of the linear position of the tense mor-
pheme in the main clause: core-medial tense indicates declarative IF, as in
Figure 1.3; tense before the first constituent of the core signals interrogative IF
(e.g. Did Mary eat a cookie after lunch?); and the lack of tense in the main clause
indicates imperative IF. This way of signalling IF is relatively uncommon
cross-linguistically. More common and straightforward is the Lakhota
approach, which involves particles to mark IF. If one wants to change (11a)
into a question, it is only necessary to add he, the question particle, to the end
of the sentence, as in (15a). If one wants to tell someone ‘Don’t dance!’, then
the imperative final particle would be added, as in (15b).
(15) a. Wíŋyaŋ kiŋ hé wačhí-kte-šni he? Lakhota

woman the that dance-pot-neg int
‘Will that woman not dance?’

b. Wačhí-šni yo/ye!
dance-neg imp.Male/imp.Female
‘Don’t dance!’
Lakhota IF particles index the sex of the speaker (Ullrich 2011: 820–822):
those ending in -o are for male speakers, while those ending in -e are for
female speakers. He in (15a) was originally a female form but has come to be
used by both sexes; the male form is huwó. The base form for the assertion
marker for men is yeló, while women use ye; kštó is used by both sexes but
primarily by women (Ullrich 2011).
These morphemes raise an important issue with respect to the concreteness
restriction on syntactic representations in RRG. The interrogative and impera-
tive particles are obligatory in questions and commands, because without
them the utterance would not be interpreted as having those IF values. The
assertion particles are different; if they are not used, as in (11a), the utterance
is still interpreted as an assertion due to the lack of the interrogative and
imperative markers. Hence there are two ways to indicate an assertion in
Lakhota: use of an overt assertion marker, or use of no IF particle of any kind.
The first is unproblematic: there is an overt form, kštó for example, which can
be linked to the operator projection as the IF operator with scope over the
clause layer. The problem arises in the second case in which there is no IF
marking of any kind, yet the sentence has a definite IF value. It is necessary to
represent this, especially in the analysis of complex sentences; this will be
discussed in Section 1.3. This is illustrated by (16), the structure of which is
given in Figure 1.7; only the operator projections are represented.
(16) Wíŋyaŋ kiŋ hé wačhí-šni kštó/Ø Lakhota

woman the that dance-neg decl.Female/decl
‘That woman does/did not dance.’16
The issue with the second representation is whether it violates the prohib-
ition against phonologically null elements in the syntax. There are good
grounds for concluding it does not. First, the proscribed null elements are
lexical categories or projections thereof; for example, null pronominals or
verbs are considered to be members of the lexical categories of pronominal
and verb.
The zero morpheme in (16) is a closed-class grammatical (function) mor-
pheme and is not a free form which can be manipulated in the syntax,
Figure 1.7 IF indicators in (16)

unlike null pronominals or verbs. It can be viewed as part of a paradigm of IF

morphemes.17
(17) Lakhota IF morphemes: declarative interrogative imperative

Male speaker: yeló huwó yo
Female speaker: ye ~ kštó he ye
Neutral: Ø he —
Zero morphemes which are part of a paradigm of closed-class grammatical

categories do not violate the prohibition on phonologically null elements in
the syntax, which applies only to phonologically null lexical items.
1.2.3.1 The Order of Morphemes Expressing Operators and the LSC

One of the implications of treating operators as modifying different layers of
the LSC is that the linear order of the morphemes expressing the operators
reflects their scope in the LSC. When the relevant morphemes are on the
same side of the predicator in the nucleus, those expressing nuclear oper-
ators will be closer to the nucleus than those expressing core operators, and
they in turn will be closer to the nucleus than those expressing clausal
operators. This hypothesis was first put forward in Foley and Van Valin
(1984). In right-branching languages, these morphemes will typically be
realized by prefixes, whereas in left-branching languages they are usually
realized by suffixes. This is illustrated in Figure 1.8.
Because aspect is a nuclear operator and tense a clausal operator, when
they both occur on the same side of the nucleus, the morpheme expressing
the aspect operator occurs closer to the nucleus than the morpheme express-
ing the tense operator, and this is what is found in the four languages in
Figure 1.8.18 This generalization is very robust, and accordingly there are
very few exceptions to it. If in a language tense is a prefix and aspect a suffix
(or vice versa), this is not an exception or counterexample to this claim,
Figure 1.8 Examples of the ordering of aspect and tense markers in different languages

because no ordering relation between the two can be established; hence the
ordering constraint does not apply in this case.
This constraint applies to the ordering of operators modifying different
layers of the LSC, but it does not restrict the order of multiple operator
morphemes in a given layer. At the nuclear and core layers the possibilities
are limited, namely, aspect and negation at the nuclear layer, and negation
and deontic modality at the core level. With respect to aspect and negation,
it would seem that the only logical possibility is negation > aspect; in a
system with perfective and imperfective aspect, NEG [V-(im)perfective] is
meaningful (i.e. ‘not [(in)complete state of affairs]’, e.g. ‘not [finished
raining]’), but the other possibility is not (i.e. ‘(in)complete [NEG state of
affairs]’, e.g. ‘finished [not raining]’). Van Valin and LaPolla (1997: 46) give an
example of the interaction of aspect and negation from Qiang, a Tibeto-
Burman language, and as predicted, negation has scope over aspect, and the
order is negation-aspect-V.
(18) ɦɑ-m -tɕi-qɑ

e Qiang
down-neg-asp-go.1sg
‘I haven’t gone yet.’
At the core level, on the other hand, both scopes are possible. In a sentence
like John must not leave the door open, the deontic modal can have scope over
negation (i.e. ‘John is obliged not to leave the door open’), or the deontic modal
can be in the scope of negation (i.e. ‘John is not obliged to leave the door open’).
Interestingly, the second reading is stronger if the negation is contracted, as in
John mustn’t leave the door open. This variation in scope is not reflected in the
order of operator morphemes in a simple sentence, unlike in the paraphrase
which contains two cores (see Section 1.3). It is, rather, signalled prosodically.
The two readings can be distinguished clearly by different prosodic chunking
of the sentence: {John must}{not leave the door open} vs. {John must not}{leave the
door open}. The negative morpheme in English makes up for its fixed position
by being able to scope in either direction.
Multiple operators regularly co-occur at the clause level, as in Figure 1.6
in which tense and IF co-occur. There are four clausal operators (IF,
evidentials, status (clausal negation, epistemic modality) and tense), and
they fall into two groups: tense and status, on the one hand, and eviden-
tials and IF, on the other. Tense and status are propositional modifiers and
occur in both main and subordinate clauses. Evidentials and IF are utter-
ance modifiers and are restricted to main clauses. In agglutinative lan-
guages in which operator morphemes occur in sequence, as in Lakhota,
Kewa, Turkish, Tiwi and Qiang, the IF morpheme occurs farthest from the
nucleus, if it has a fixed position; this is illustrated in the Lakhota
examples in (15) and (16). In terms of the LSC operator projection, the IF
and evidential operators modify the clause node which is immediately
dominated by the sentence node; there is no such restriction on tense or
status, hence they can occur freely in subordinate clauses. There is no

scope asymmetry between tense and status; they modify at the same level,
but because the tense and status morphemes must be linearized, lan-
guages show variation in the order of these morphemes.
1.2.3.2 Clausal Operators and the PrDP and PrCS

The important distinction between the PrDP and the PrCS was character-
ized in terms of two primary differences: the PrDP is set off by an inton-
ation break, and if the XP in the PrDP is a semantic argument of the
predicator in the nucleus, then there can, and in some languages, must
be a resumptive pronoun in the core, whereas the PrCS has neither of these
features. The explanation for this opposition can be found in Figure 1.6 in
the interaction between the constituent and operator projections. These
differences follow from the fact that the PrCS is inside the clause and
therefore in the scope of the IF operator, whereas the PrDP is outside of
the clause and consequently is not in the scope of the IF operator. Being
within the scope of the IF operator means that the XP in that position can
be asserted, questioned or negated, possibilities denied to XPs outside the
scope of the IF operator. With respect to wh-question formation, the
following contrast is quite clear.
(19) a. [S [CL [PrCS What ] {did} [CORE Sally [NUC borrow ] from Bill? CORE] CL] S]
b. [S [PrDP *As for whati ][CL{did} [CORE Sally [NUC borrow ] iti from Bill? CORE] CL] S]
In (19a) what is in the PrCS, which is clause-internal and therefore in the

scope of the IF operator. There is a single intonation contour over the clause,
and there is no break after what. There is no need for a resumptive pronoun,
because all of the arguments of the predicator are in the clause. The situation
is very different in (19b), which is ungrammatical. Here the wh-expression
occurs outside of the clause and therefore outside of the domain of the IF
operator. It is not part of the speech act associated with this sentence. It cannot
be interpreted as the focus of the utterance, unlike what in the PrCS in (19a).
The intonation break following the PrDP is related to the IF operator. The
intonation contour of the primary speech act begins at the first element in
the clause19 and ends at the end of the clause. The PrDP is not part of the
clause and accordingly is not part of the intonation contour. It is a prosodic
unit of its own, and the intonation break is the transition from the prosodic
contour of the PrDP and that of the clause.
The usual test for whether an XP is asserted or not is the denial test with
only the new information given, as in (20).
(20) Speaker 1: Mary bought a new Tesla.

Speaker 2: No, a Mercedes.
This felicitous pair of utterances shows that the RP a new Tesla can be the
focus of the assertion and therefore denied. This is not possible with an RP in
the PrDP, as in (21a), but is fine with an XP in its default position, as in (21b),
or in the PrCS, as in (21c).

(21) a. Speaker 1: On Thursday, Mary will meet the journalist at the library.
Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/??No, Friday
b. Speaker 1: Mary will meet the journalist at the library on Thursday.
Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/No, Friday
c. Speaker 1: Thursday Mary will meet the journalist at the library.
The same restrictions hold with respect to forming questions.
(22) a. Speaker 1: On Thursday, will Mary meet the journalist at the library?
Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/??No, Friday
b. Speaker 1: Will Mary meet the journalist at the library on Thursday.
c. Speaker 1: Thursday will Mary meet the journalist at the library.
Thus, the features distinguishing the PrDP and the PrCS are directly related
to the fact that the scope of the IF operator is the clause. The same holds for
the PoDP and PoCS; as the Dhivehi examples in (6) show, the PoCS can host wh-
expressions analogous to (19a) and contrastive focus phrases analogous to (4b).
The PoDP, like the PrDP, cannot be asserted or questioned, as illustrated in
(23a, b), nor can it host wh-expressions, as shown in (24).
(23) a. Speaker 1: Mary bought them ice cream cones, the kids.
Speaker 2: No, some popcorn/No, Bill did/??No, the adults.
b. Speaker 1: Did Mary buy them ice cream cones, the kids?
Speaker 2: No, some popcorn/No, Bill did/??No, the adults.
(24) a. *Mary bought them ice cream cones, who?

a′. *Did Mary buy them ice cream cones, who?
b. *Mary bought it for the kids, what?
b′. *Did Mary buy it for the kids, what?
Thus, recognizing the IF operator as having clausal scope leads to a deeper

understanding of the differences between the PrDP and PoDP, on the one
hand, and the PrCS and PoCS on the other.
1.2.3.3 The Status of Copular Verbs Like English be

Copular verbs like English be present interesting issues regarding their repre-
sentation in the LSC. In the Lakhota examples (11b), (13b) and (14) there is no
copula of the kind required in their English translation. This suggests that in
such constructions be is not the predicator in the nucleus in English but
rather is just an auxiliary accompanying the actual predicator, namely,
woman or tall. The same holds for the Tagalog example in (10b): bata ‘child’
is the predicator in (10b) and its English translation. Thus, this use of the
English copula is not found in many other languages, and it would be odd to
say that in English be is the predicator in these examples, but in Tagalog and
Lakhota the predicator is the nominal or stative verb. Rather, the predicator
in the constructions in all three languages is ‘woman’, ‘child’ or ‘tall’, and
English requires the auxiliary be when the predicator is non-verbal.

Figure 1.9 Passive be in (25a) and progressive be and passive be in (25b)
There is another construction in English in which auxiliary be is obliga-

tory, namely, the passive construction. It is possible to have two be auxiliar-
ies in the same sentence, as in (25b), but the representation of the two is
quite different, as can be seen in Figure 1.9.
(25) a. Max was interviewed by CNN during the insurrection.

b. Max was being interviewed by CNN during the insurrection.
In (25a), the tense operator is realized on the passive auxiliary be, and
its core-internal linear position marks declarative IF. In this construction
was is attached to both projections: the tense and IF operators connect it
to the operator projection, while the be to which these operators are
attached is structurally required for passive nucleus formation.20 In
(25b), the auxiliary was is attached to the operator projection three times:
(1) it is part of the two-part progressive aspect morpheme, beþ-ing; (2) it
signals past tense; and (3) its core-internal position in the linear string
indicates declarative IF. Passive be is not connected to the operator pro-
jection; it hosts -ing, which is part of the expression of progressive aspect
and is attached to the operator projection. Thus, passive be is always part
of the constituent projection and may also be connected to the operator
projection, whereas progressive be can only be connected to the operator
projection.
This follows from the fact that progressive be occurs only when the
progressive aspect is expressed. Passive be, like copular be, is structurally
required for proper nucleus formation and occurs even when there are no
operators, as in the infinitives in (26).
(26) a. Max wanted/tried to be interviewed by CNN during the insurrection.

b. Max wants to be stronger/a lawyer.

It should be noted that there are other auxiliary verbs which can fulfil be’s
structural role while contributing semantically to the predication. For
clauses with adjectival predicates, seem and appear can replace be as in Bill
is/seems/appears happy/stronger/healthy. Become and get can also serve this func-
tion: Mary became despondent, Sally got stronger. Get can also replace be in a
passive construction, as in Max got interviewed by CNN during the insurrection.
1.2.3.4 Universal Operators

Of all the operators discussed in RRG, only two are universal, found in every
language: negation and IF. In every human language it is possible to make
assertions, ask questions and give commands. That is fundamental to lan-
guage as a system of communication. Hence IF is an operator in every
language. This is not to say that the inventory of speech acts is exactly the
same in every language; rather, it is a claim that in every language it is
possible to make the three essential speech acts: asserting, questioning,
commanding.
Negation is essential for both communication and reasoning. With respect
to the former, denials (e.g. ‘X is not the case’, ‘X did not do Y’) and negative
commands (e.g. ‘Don’t do Z’) are essential and very important types of
utterances. With respect to the latter, practical reasoning involves negation
(e.g., ‘if I do X, I will not get Z, but if I do Y, I will get Z’). The exact detail of
how negation works is not necessarily the same in every language, for
instance a language may or may not have negative polarity items, but
negation is part of the system in every language.
A distinctive property of negation is that it is the sole operator that can
modify all of the layers of the LSC: nuclear negation has just the nucleus in its
scope, core negation has the core or some subpart, typically a core argument,
in its scope, and clausal negation has the whole clause in its scope. One might
expect that languages would typically have three negative morphemes, one
for nuclear negation, one for core negation, and one for clausal negation, but
this is in fact quite rare. Some languages have two negative morphemes with
different scopes; for example, Barai, a Papuan language, has a negative
morpheme ba which is the only element that can occur after the verb in
Barai and has clausal scope (external negation), and another one, naebe, which
occurs before the verb and can have nuclear or core scope (internal negation)
(Foley and Van Valin 1984: 192–193). Korean also has two negative construc-
tions, one which has nuclear scope, as shown in (27a), and one with core or
clausal scope, as in (27b), from Yang (1994: 127).
(27) a. Chelswu-ka pap-lul manhi an - mek-ess-ta.

-nom dinner-acc much neg-eat-pst-decl
‘Chulsoo did not eat much dinner (But he ate a little).’
b. Chelswu-ka pap-lul manhi mek-ci.ahn-ess-ta
-nom dinner-acc much eat-neg-pst-decl
‘It is not the case that Chulsoo ate much dinner,’ ‘Chulsoo did not eat
much dinner .’

The negative morpheme is an-/ahn-, and when it is prefixed to the verb, as in

(27a), it has nuclear scope, but when it is suffixed to the verb, as in (27b), it
may be interpreted as having core or clausal scope. Both Barai and Korean
have two negative constructions, but the opposition in Barai is clausal vs.
non-clausal, while in Korean it is nuclear vs. non-nuclear.
The usual case is like that in English and other languages with a single
negative in a fixed position with an ambiguous interpretation.
(28) Max didn’t buy a new car.

a. He leased one. Nuclear scope
b. He just went window shopping Core scope
c. He bought a new motorcycle. Core (argument) scope
d. Bill did. Core (argument) scope
e. It’s not the case that Max bought anything. Clausal scope
The scope of negation is tied to the focus structure of the clause, and
consequently the interpretation depends on the context in which the
sentence occurs.
1.2.4 Peripheries
In the earlier versions of the LSC there was only a single periphery, which
was attached at the core level. This proved to be wholly inadequate, and in
Van Valin (2005) it was proposed that each of the layers of the spine of the
LSC was modified by a periphery containing modifiers appropriate to the
layer. Moreover, a distinction was made between phrasal and non-phrasal
adjuncts, phrasal adjuncts being PPs and non-phrasal adjuncts being
adverbs, in simple sentences.21 In complex sentences, phrasal adjuncts also
include what are traditionally called adverbial clauses (to be discussed in
Section 1.3).
1.2.4.1 Phrasal Adjuncts

The two types of adjuncts are subject to different ordering constraints.
Languages tend to have restrictions regarding the placement of phrasal
adjuncts. In English, phrasal adjuncts occur after the core, as in (3), whereas
in languages with primarily nucleus-final word order, they appear after the
subject and before the remaining core arguments, for example Japanese.
(29) Hanako ga tosyokan de hon o yon-da.

nom library in book acc read-pst
‘Hanako read a book in the library.’
The adjunct PP tosyokan de ‘in the library’ is linearly within the core, and the
LSC diagram necessarily involves crossing tree branches, as in Figure 1.10.
The PP tosyokan de ‘in the library’ is in the core-level periphery, because it
expresses the location of the event described by the core, that is, Hanako’s
reading a book, and consequently the representation of the hierarchical
structure and the linear sequence of words leads to the crossing branches.

Figure 1.10 Default placement of phrasal adjuncts in Japanese
Such crossing branches are allowed in RRG, because the tree represents
semantic dependencies primarily rather than purely syntactic constituent
structure.
German presents an interesting case, as there has long been a controversy
as to whether it is fundamentally verb-medial or verb-final. With respect to
the default placement of adjunct PPs, it patterns with Japanese and not with
English.
(30) a. Max hat nach der Auseinandersetzung mit Maria eine ganze Flasche Wein getrunken.
has after the argument with a entire bottle wine drunk
‘Max drank an entire bottle of wine after the argument with Mary.’
b. Max trank nach der Auseinandersetzung mit Maria eine ganze Flasche Wein.
drank after the argument with a entire bottle wine
‘Max drank an entire bottle of wine after the argument with Mary.’
In (30a) the nucleus is core- (and clause-) final, as in Japanese, and the order
of arguments and adjuncts is the same, too: subject, core-level adjunct PP,
direct object. In (30b) the nucleus is in second position in the core, as in
English, but the default order of arguments and adjuncts is the same as in
(30a) and Japanese. On the other hand, the word order in (30b) is impossible
in English; nothing is allowed to occur between the nucleus and the
following direct core argument (‘direct object’).22
Phrasal adjuncts may also appear in positions other than their default
position. In English and many other languages, they may appear in the PrDP
and function as a frame-setting topic.
(31) a. After the party, Sam went home and watched TV for a while.
a′. After the party, where did Sam go?
b. Because of the cold weather, vaccine distribution has been slowed down
significantly.
b′. Because of the cold weather, has vaccine distribution been adversely affected?
The adjunct PP in the PrDP does not express the topic of the sentence, unlike
in Figure 1.4 or (7), because (31a) is not about the party but rather about Sam
and (31b) is not about the cold weather. Rather, after the party and because of
the cold weather set the frame of reference in terms of which the following

clause(s) should be interpreted. In the questions the initial PP is not in the

scope of the IF operator, hence is outside the clause in the PrDP.
It is also possible for adjunct PPs to occur in the PrCS as contrastive frame-
setting topics, as in (32).
(32) a. At the mall Sally couldn’t find what she wanted, but in a small boutique
she found it.
b. Before lunch Bill drank only coffee, but after lunch he ate a lot of snacks.
These initial PPs are not set off prosodically like the ones in (31) and are part
of the assertion, that is, they are within the scope of the IF operator, and are
therefore in the PrCS. This entails that they do not occur in question,
because the PP is in the actual focus domain (AFD; see Chapter 11).
(33) a. ??Before lunch did Bill drink only coffee, but after lunch did he eat a lot
of snacks?
b. *Before lunch what did Bill drink, but after lunch what did he eat?
Both examples are fine if there is a pause after the initial PPs, which
indicates that the structure is the same as in (31), not (32).
It was mentioned with respect to (3) in Figure 1.3 that the order of adjunct
PPs is constrained by the layer they modify, with core-level modifiers preced-
ing clause-level modifiers in a right-branching language like English. When
there are multiple adjunct PPs which modify at the same level, however, the
order is ‘free’ and influenced by information structure, as in (35).
(34) a. Mary ate a cookie after lunch despite her diet. [¼(3)]
a′. ??Mary ate a cookie despite her diet after lunch.
b. Sally met Max in the library because of the cold weather.
b′. ??Sally met Max because of the cold weather in the library.
(35) a. Mary ate a cookie after lunch in the garden.

a′. Mary ate a cookie in the garden after lunch.
b. Sally met Max in the library after the lecture.
b′. Sally met Max after the lecture in the library.
(36) a. Vaccine distribution has been adversely affected despite the government’s
best efforts because of the cold weather.
b. Vaccine distribution has been adversely affected because of the cold
weather despite the government’s best efforts.
PPs headed by after and in are core-level modifiers, while those headed by
despite and because of are clause-level modifiers, and consequently the
ordering restriction in (34) and the lack thereof in (35) and (36) follows from
the analysis of phrasal adjuncts in terms of the LSC.
1.2.4.2 Non-Phrasal Adjuncts

The class of non-phrasal adjuncts consists primarily of adverbs, and they
have the same possibilities of occurrence as phrasal adjuncts and then some,
as in (37).

(37) a. Mary quietly ate a cookie after lunch in the garden despite her diet.
b. Mary ate a cookie quietly after lunch in the garden despite her diet.
c. Mary ate a cookie after lunch quietly in the garden despite her diet.
d. Mary ate a cookie after lunch in the garden quietly despite her diet.
e. ??Mary ate a cookie after lunch in the garden despite her diet quietly.
f. Quietly(,) Mary ate a cookie after lunch in the garden despite her diet.
Quietly is a type of manner adverb and is a core-level modifier, which

accounts for the oddity of it following the clause-level adjunct PP in (37e),
just as in (34a′, b′). If it is replaced by a clause-level adverb (e.g. evidently), the
situation is somewhat different.
(38) a. Mary evidently ate a cookie after lunch in the garden despite her diet.
b. ?Mary ate a cookie evidently after lunch in the garden despite her diet.
c. ?Mary ate a cookie after lunch evidently in the garden despite her diet.
d. Mary ate a cookie after lunch in the garden evidently despite her diet.
e. Mary ate a cookie after lunch in the garden despite her diet evidently.
f. Evidently, Mary ate a cookie after lunch in the garden despite her diet.
The oddity of (38b) and (38c) results from having a clause-level modifier
occurring in the midst of core-level modifiers. There is a striking contrast in
acceptability between (37e) and (38e), which is a function of the difference in
the level of modification (core vs. clausal) of the adverbs. The two ordering
possibilities in (38d, e) show that it makes no difference with respect to the
sequence whether the adjuncts are phrasal or not, as long as they are
modifiers of the same layer.23
As (37) and (38) make clear, adverbs can occur in positions which are not
part of the default periphery for phrasal adjuncts, and it was proposed in
Van Valin (2005) that all adjuncts not in the default periphery, the Pr/PoDP
or the Pr/PoCS are in a periphery specific to them. This is exemplified in
Figure 1.11.
The LSC is crucial for the analysis of the scope of modification of both
phrasal and non-phrasal adjuncts, but it has significant limitations, namely,
Figure 1.11 Periphery for individual adverb in (37a)

there are only three layers but there are many more types of adverbs. This is
illustrated in (39); the list of types of adverbs is not exhaustive.
(39) Adverbs and the LSC

a. Clausal modifiers:
1. Speech act modifiers: candidly, honestly, . . .
2. Evidential: evidently, allegedly, . . .
3. Speaker attitude/judgement: appallingly, unfortunately, . . .
b. Core modifiers:
1. Temporal: yesterday, tomorrow, . . .
2. Manner
a. Pace: quickly, slowly, . . .
b. Intent: deliberately, intentionally, . . .
c. Performance: carefully, sloppily, . . .
d. Action: violently, weakly, . . .
c. Nuclear modifiers:
1. Aspectual: completely, continuously, . . .
Two points deserve mention. First, a given adverb can be used in more than
one category, as in the well-known example of cleverly as in Mary hid the
jewellery cleverly (manner-performance) vs. Cleverly, Mary hid the jewellery
(speaker attitude/judgement) vs. Mary cleverly hid the jewellery (ambiguous).
Second, even though the semantic domains of operators and adverbs are
similar, adverbs are not operators; adverbs are open-class lexical items,
whereas operators are closed-class grammatical elements. This is particu-
larly important when it comes to distinguishing the operator tense and
temporal adverbs; the former is a clausal operator, whereas the latter is a
core-level modifier. This can be seen clearly in sentences like (40).
(40) [CLAUSE [CORE1 Fred told Sally][ yesterday] [CORE2 to meet him][ in Paris Friday]].
This sentence consists of a single clause with a single tense operator which is
made up of two cores, each of which is modified by a different temporal
adverb. Each temporal adverb modifies the core it occurs with, while the
tense operator has scope over the entire clause.
It is reasonable to speculate that the largest class of adverbs is manner
adverbs, which has a number of subtypes, all of which are core-level modi-
fiers. Despite them all being core modifiers, the ordering among them does
not appear to be unconstrained.
(41) a. Mary will intentionally speak more clearly tomorrow.
b. ?Mary will tomorrow speak more clearly intentionally.
c. Mary will speak more clearly intentionally tomorrow.
d. Mary will speak intentionally more clearly tomorrow.
e. *Mary will more clearly speak intentionally tomorrow.
f. ?Mary will intentionally speak tomorrow more clearly.
g. ?Mary will tomorrow speak intentionally more clearly.
There are two manner adverbs, intentionally (intent) and clearly (perform-
ance), and one temporal adverb, tomorrow, in these sentences. Clearly seems

to be the most constrained, because it is strongly preferred after the verb,

whereas intentionally can appear either before or after the verb. Tomorrow is
best in final position, and it is the only one of the three that can appear in
initial position without changing its meaning; intentionally can’t appear at
all initially, and clearly is interpreted as an expression of speaker attitude/
judgement rather than manner.
(42) a. Tomorrow(,) Mary will intentionally speak more clearly.

b. Clearly, Mary will intentionally speak more intelligibly tomorrow.
c. *Intentionally, Mary will speak more clearly tomorrow.
Thus, core-modifying adverbs differ in terms of their scope of modification

within the core. Temporal adverbials appear to have the widest scope, and
the relative scope relations among the manner adverbs remain to be worked
out. This situation is reminiscent of the constraints on the ordering of
adjectives, all of which modify the same level of the RP (see Section 1.2.5).
(43) a. the expensive big red hand-made scarf

b. *the big hand-made red expensive scarf
c. *the expensive hand-made red big scarf
d. *the red expensive big hand-made scarf
1.2.4.3 Extended Peripheries

It was pointed out in Section 1.2.4.1 that right-branching languages like
English tend strongly to have the default periphery for phrasal adjuncts, PPs
and clauses, after the core, while left-branching languages like Japanese and
German have it after the subject and before the direct object and the
nucleus.24 It has been discovered that there are instances in which the
periphery for phrasal adjuncts gets extended to before the nucleus in
right-branching languages. Examples can be found in Italian (Bentley 2008:
278) and in Tagalog (Latrouite and Van Valin 2021). The example from Italian
is given in (44).
(44) a. Il libro, per sbaglio, Luca ha strappato, non il quaderno.

the book by mistake has torn not the workbook
‘It is the book that Luca tore by mistake, not the workbook.’
b. [[[PrCS Il libro], [PERIPH per sbaglio], [CORE Luca ha strappato]], [PoDP non il
quaderno]]
In (44) there is an adjunct PP, per sbaglio ‘by mistake’, which appears
between the PrCS RP il libro ‘the book’ and the core-initial subject Luca.
Normally, phrasal adjuncts occur after the core, but in this case, the adjunct
PP is in its own periphery, analogous to the non-phrasal adjunct in
Figure 1.11.
The Tagalog example is from Schachter and Otanes (1972: 498–499); it
involves a construction they label ‘non-emphatic inversion’, in which an
adverbial is displaced from its normal post-core position to the beginning of
the clause.

(45) a. P<in>asok nila ang silid nang bigla.

<uv.pfv>enter 3pl.gen nom room lnk suddenly
‘They suddenly entered the room’, ‘They entered the room suddenly.’
b. Bigla nila-ng p<in>asok ang silid.
suddenly 3pl.gen-lnk <uv.pfv>enter nom room
‘Suddenly, they entered the room.’
In the discussion of (5) from Tagalog, it was pointed out that the personal
pronouns are second-position clitics, and in (45b) nila ‘3plgen’ follows bigla
‘suddenly’, which shows it is in clause-initial position. It might be concluded
that it is in the PrCS, but it lacks the contrastiveness of PrCS XPs in Tagalog.
In fact, (46) shows that it is in the same place as per sbaglio in (44): between
the PrCS and the core.
(46) a. Nag-trabaho ang lahat nang mabilis kanina-ng umaga.

av.rls-work nom all lnk quickly this-lnk morning
‘Everyone worked quickly this morning.’
b. Kanina-ng umaga mabilis na nag-trabaho ang lahat.
this-lnk morning quickly lnk av.rls-work nom all
‘It was this morning that everyone worked quickly.’
c. [CLAUSE [PrCS Kanina-ga umaga][PERIPHERY mabilis-na][CORE nagtrabaho ang lahat]]
The default position for the two adjuncts is given in (46a); they both follow
the single core argument. In (46b), on the other hand, they appear before the
nucleus, with the temporal adjunct kanina-ng umaga ‘this morning’ in
clause-initial position in the PrCS, as indicated by the contrastive interpret-
ation (cf. (5b)), and the adverbial mabilis ‘quickly’ between it and the core-
initial nucleus nagtrabaho ‘worked’. A comparison with (44b) reveals that the
structures are completely parallel.
It was mentioned earlier that in German the default location for adjuncts,
both phrasal and non-phrasal, is after the subject and finite verb or auxiliary
and before the remaining elements in the core. One of the relatively unique
features of German main clause structure is what is traditionally character-
ized as the ‘prefield’, the clause-initial position before the finite verb or
auxiliary which must be obligatorily filled. Wh-expressions normally occur
in it in questions, suggesting it is the PrCS, but in many sentences, the XP in
it is not contrastive or information-structurally special in any way. In (30),
for example, Max is a normal, topical subject in the prefield. In (47) it is
occupied by an adjunct which need not have any special information-
structural status.
(47) a. Nach der Vorlesung hat Dieter in der Bibliothek Almuth getroffen.
after the lecture has in the library met
‘After the lecture Dieter met Almuth in the library.’
b. Gestern hat Dieter in der Bibliothek Almuth getroffen.
yesterday has in the library met
‘Yesterday Dieter met Almuth in the library.’

Ruhnau (2011) gives an RRG analysis of the prefield and argues that in
(47a, b) the clause-initial adjunct is a peripheral constituent, not in the
PrCS. The prefield is thus rather heterogeneous structurally; it can be a
core-argument RP, as in (30), or PP, it can be the PrCS in wh-questions and
contrastive topic or focus sentences, or it can be an adjunct in
a periphery.
1.2.5 The Structure of Phrases

The LSC is the model for the analysis of structure in RRG. Accordingly,
phrases smaller than a clause are also analysed as having a layered struc-
ture, in particular PPs, MPs and RPs.
1.2.5.1 Adpositional Phrases

The structure and function of prepositional and postposition phrases is
discussed in detail in Ibáñez Cerda (Chapter 10), and consequently the
discussion in this chapter will be brief. The most important distinction
made regarding the structure of PPs is the contrast between predicative and
non-predicative PPs, adopted from Bresnan (1982). Predicative PPs, as the
name implies, have an adpositional predicate in the nucleus of the PP;
this will be discussed in the section devoted to the semantic representation
of sentences in RRG. The object of the adposition is a core argument in the
layered structure of the adpositional phrase (LSPP), as illustrated in
Figure 1.12 with the German PPs aus dem Haus ‘out of the house’ and dem
Haus gegenüber ‘across from the house’.
German has both prepositions (e.g. aus ‘from, out of’) and postpositions
(e.g. gegenüber ‘across from’); in both structures there is an adpositional
predicate in the nucleusP, and the object is in the coreP. This is the only
endocentric structure posited in RRG.
There are a few significant differences between the LSC and the LSPP.
First, there is only one layer above the coreP, the PP layer, unlike in the
LSC with clause and sentence above the core. Second, there are no extra
positions like the Pr/PoDP or the Pr/PoCS. Third, there is no operator
projection, because there are no operators modifying PPs. Fourth, there
Figure 1.12 German predicative PPs

aren’t peripheries at every level, as in the LSC, as there are very few
lexical modifiers for PPs. The best examples are adverbs like right or
immediately, as in right behind the chair or immediately before the accident.
They would appear to be coreP-level modifiers. Predicative PPs share the
property of recursion with clauses in a limited way, as in The mouse ran out
from under the bed.
(48) Structure of the recursive PP out from under the bed

[PP[DIR out[COREp[NUCp from][COREp[NUCp under] the bed COREp]COREp] DIR] PP]
In locative inversion constructions, the whole complex PP must be inverted,

that is, Out from under the bed ran a mouse, not, for example, *Under the bed ran a
mouse out from/Under the bed ran out from a mouse, but From under the bed a mouse
ran out is fine, showing that out is a directional modifier rather than
a preposition.
Non-predicative PPs, on the other hand, do not have a full layered struc-
ture as in Figure 1.12. The adposition in the PP is not a semantic predicate,
and therefore the object is not a semantic argument of the adposition.
Rather, the adposition is functionally like a case marker, and the resulting
structure is not endocentric. In RRG non-predicative adpositions are either
assigned by rule or listed in the lexical entry of the verb. The classic example
of a non-predicative adposition is the to marking the recipient argument of a
transfer verb. This is illustrated in Figure 1.13 from English and Bulgarian;
in both Old English and Old Church Slavonic the third argument of a
transfer verb carried the dative case, and after English and Bulgarian lost
their case systems, the dative in this context was replaced by a preposition,
to in English and na in Bulgarian.
The primary content in these phrases is the RPs, and therefore these are
exocentric phrases. The adpositions are basically case markers,25 and in
English and many other languages, their lack of semantic significance is
shown by the fact that they are not obligatory, as in (49).
(49) a. Nancy gave the flowers to the woman.

a′. Nancy gave the woman the flowers.
b. Sam presented the trophy to the winner.
b′. Sam presented the winner with the trophy.
Figure 1.13 Non-predicative PPs in English and Bulgarian

Other examples of alternating non-predicative prepositions can be found

in (50).
(50) a. Bill loaded the vehicle with grapes.
a′. Bill loaded the grapes into the vehicle.
b. Max drained the water from the swimming pool.
b′. Max drained the swimming pool of its water.
c. Mary sprayed paint on the wall.
c′. Mary sprayed the wall with paint.
In some cases each member of the alternation occurs with a different verb,
as in The gang stole $5000 from the bank vs. The gang robbed the bank of $5000.
Non-predicative adpositions cannot be stacked the way predicative adposi-
tions can be, as in the earlier example out from under the bed.
The distinction between predicative and non-predicative adpositions is
very important for the analysis of many grammatical phenomena cross-
linguistically.
1.2.5.2 Modifier Phrases

Canonically, nouns are modified by adjectives and nouns, while verbs and
adjectives are modified by adverbs, but this is not always the case. The
examples in (12) are repeated here, as they do not fit the traditional scheme.
(12) a. The God is dead philosophers are mostly dead.

b. My grandson likes to give me the who’s the boss now, silly old grandpa wink
frequently.
In (12a) the modifier in the ‘adjective slot’ is a clause, and in (12b) it is a

sentence complete with PrCS and PoDP. That they are in an ‘adjective slot’
and are not some kind of special relative clause or the like is shown by the
fact that they may co-occur with regular adjectives in a sequence of modi-
fiers, as in (12′).
(12′) a. The notorious [God is dead] continental philosophers are mostly dead.
b. My grandson likes to give me the triumphant [who’s the boss now, silly old
grandpa] sly wink frequently.
There are other examples of phrasal modifiers in a position which is

reserved for adjectival heads: English *the very proud of her daughter woman
is impossible, but it is possible in German, die auf ihre Tochter sehr stolze Frau
[the on her daughter very proud woman] ‘the woman very proud of her
daughter’. MPs account for these modifying expressions, since they are non-
endocentric like RPs and nuclei. MPs occur in the peripheries of the element
modified; they are not involved in predicative uses of nouns, adjectives
and adverbs.
The German example is important, because it shows that MPs have a
layered structure like other phrases. The nucleus of the MP is the adjective
stolz ‘proud’, which takes a single core argument, the PP auf ihre Tochter ‘of
[on] her daughter’. The structure of the whole RP die auf ihre Tochter sehr stolze

Figure 1.14 The layered structure of the MP
Frau ‘the woman very proud of her daughter’ is presented in Figure 1.14. Two
points need to be mentioned. First, degree modifiers like sehr ‘very’ do not
head phrases and therefore are not in an MP. Second, the structure of RPs,
including the status of definite articles like die ‘the’ will be discussed in the
next section.
Like the predicative PP, the MP has one less layer than the LSC, and there
are no operators modifying it. There are peripheries at the nuclear level for
degree modifiers (illustrated) (see also Fleischhauer 2016) and at the core
level for manner adverbial modifiers, for instance the rapidly contained wild-
fire, the beautifully embroidered dress, the carefully constructed maze.
In the English translation of this RP, the MP, which is a reduced relative
clause, follows the head noun; when the MP consists of a participle, some,
but not all, may occur either pre-head or post-head, for example the murdered
man vs. the man murdered, the running man vs. the man running, the embroidered
sweater vs. *the sweater embroidered (but the sweater embroidered by Mary).
It was pointed out that the unusual phrasal modifiers in (12) cannot be
considered some kind of exotic relative clause or the like, as the examples in
(12′) show. What is so unusual about them is that they exhibit asymmetrical
embedding, that is, a larger unit (clause, sentence) is embedded in a smaller
unit (nucleus); this will be discussed in Section 1.3. The structure of the first
RP in (12a) is given in Figure 1.15; the structure of the RP with the sentence
as modifier in (12b) can be found in Van Valin (2008: 173).
The fact that the embedded clause is under the nucleus rather than being
under the MP node follows from the examples in (12′), in which the phrasal
modifier occurs in the middle of a series of modifiers. The structure of RPs
with multiple modifiers, for example the very expensive beautifully embroidered

Figure 1.15 The structure of the RP in (12a)
red sweater, involves multiple sequentially embedded coreM nodes, as illus-

trated in Figure 1.16.
The recurring embedded units are coreM nodes for the reason given
earlier: the modifying expression can be modified by manner adverb, which
is a core-level modifier.
1.2.5.3 Reference Phrases

RPs have a more complex layered structure than PPs and MPs. They have
operators, one of which, definiteness, is represented in Figures 1.14–1.16.
They have peripheries at three levels, nucR, coreR, and RP, and they have
optional positions analogous to the Pr/PoCS and the Pr/PoDP. However, like
PPs and MPs, the layered structure of RPs has one layer less than the clause,
as is clear in the figures mentioned above.
As with the operators modifying the layers of the LSC, the RP operators are
closed-class, grammatical elements. The operators for each layer are associ-
ated with a specific semantic domain: the nuclearR operators express quali-
tative features of the referent, the coreR operators express quantitative
characteristics of the referent, and the RP-level operators locate the referent
within the immediate common ground, which includes the discourse

Figure 1.16 Structure of RP with multiple modifiers
context and the physical environment. These distinctions derive from the
proposals in Rijkhoff (1990, 2002). The primary nuclear operator in the
clause is aspect, which is realized prototypically by the imperfective–
perfective opposition, and a number of linguists, for example Jackendoff
(1990) and Talmy (2000), have argued that the imperfective–perfective con-
trast for verbs has, as its analogue among nouns, the count–mass distinc-
tion, with mass and imperfective, on the one hand, and count and
perfective, on the other, going together. If one needs to render a mass noun
(e.g. water) countable, languages make use of measure phrases or numeral
classifiers, for example two glasses/bottles/drops of water. A related use of
nominal classifiers is to specify the relevant semantic properties of a refer-
ent. Becker (1975), for example, gives the following Burmese examples of
classifiers contributing to the interpretation of an RP.
(51) a. myiɂ t yaɂ
e ‘river one place’ (e.g. destination for a picnic)
river one CL
b. myiɂ t tan
e ‘river one line’ (e.g. on a map)
c. myiɂ t hmwa
e ‘river one section’ (e.g. a fishing area)
d. myiɂ t ′sin
e ‘river one distant arc’ (e.g. a path to the sea)
e. myiɂ t θwɛ
e ‘river one connection’ (e.g. tying two villages)
f. myiɂ t ′pa
e ‘river one sacred object’ (e.g. in mythology)
g, myiɂ t khu′
e ‘river one conceptual unit’ (e.g. in a discussion of rivers
in general)
h. myiɂ t myiɂ
e ‘river one river’ (the unmarked case)

This nuclearR operator will be termed nominal aspect, and it covers the
count–mass distinction and the use of classifiers, numeral or nominal.
The core-level operators in the RP are concerned with quantity and quan-
tification. The primary operator notion at this level is number. It is unques-
tionably closed class and grammatical. Less clear-cut is whether numerals
and quantifiers are operators belonging to this domain. In previous work in
RRG, numerals were assumed to be coreR operators, but there are significant
problems with that analysis. First, numerals are not a closed class; in fact,
they are an infinite open class. Second, grammatically they behave like
lexical items, not grammatical morphemes. For example, they can function
as arguments in elliptical contexts and sometimes as part of the predicator
in the nucleus; number cannot.
(52) a. I’ll take two.

b. Q: How many people will participate?
A: Seven should./The expected number is seven.
Quantifiers present a range of behaviour, and this makes it impossible to

give the whole class a uniform, consistent analysis. There are those that act
like lexical items, as well as modifiers (e.g. Our problems are many, our options
are few vs. We have many problems and few options). There are pairs of quanti-
fiers which are very close in terms of meaning but which have quite
different grammatical behaviour. For example, each and every are such a
pair; each dog barked and every dog barked are both fine, but only each can
stand alone in an elliptical context, for instance What did the dogs do? Each
barked/*Every barked vs. Each/every one barked. They are a closed class but share
important properties with lexical items. Accordingly, they should be ana-
lysed as lexical modifiers at the coreR-level, together with numerals. Hence
grammatical number is a coreR operator and is linked to the operator
projection of the RP, while the other two categories are lexical modifiers
which occur in the coreR-level periphery. The other possibility for analysis
as a coreR-level operator is negation, as in no book or its German counterpart
kein Buch. Given that negation is uncontroversially an operator in the
clause, it should be analysed as one in the RP. As in the clause, negation
can modify at more than one level in the RP. NuclearR negation is expressed
by no, e.g. no (brown) dogs were seen today, as it takes scope only over bare
nouns or adjective þ noun. It cannot have scope over quantifiers or RP-level
modifiers, e.g. demonstratives; not must be used (e.g. *no every student vs. not
every student, *no that book vs. not that book). Interestingly, no can take
numerals in its scope, as in no two snowflakes are alike, which suggests that
numerals are more akin to adjectives (nuclearR-level modifiers) than to
quantifiers (coreR-level modifiers).
The third layer for operators is the RP level, which locates the referent of
the RP in the immediate common ground (Berio et al. 2017). The immediate
common ground consists of the physical and social environment in which
the interaction takes place and the discourse context. The two notions that

are relevant here are definiteness and deixis. Definiteness covers several
related oppositions: referential vs. non-referential, specific vs. non-specific,
and definite vs. indefinite. These distinctions are canonically expressed by
articles, but it is well known that many languages lack articles. Neverthe-
less, these notions are relevant in article-less languages and are expressed by
word order, demonstratives, and/or the numeral meaning ‘one’ (Balogh et al.
2020), for example. Lakhota has a particularly rich system of articles, with
two definite articles, and an astonishing number of indefinite articles (see
Ullrich 2011: 810–815). The central distinctions are illustrated in the
following dialogue (Ullrich and Black Bear 2018: 108).
(53) a. Okíčhize olówaŋ waŋží yéksuya he?

War song art you.remember q
‘Do you remember a war song?’
b. Haŋ, okíčhize olówaŋ waŋ wéksuye.
yes war song art I.remember
‘Yes, I remember a (certain) war song.’
c. Olówaŋ kiŋ líla onáȟ’uŋ wašté.
song art very good.to.listen.to
‘The [war] song is really beautiful’
d. Yéksuya he? John Vietnam etáŋhaŋ glí yuŋkȟáŋ
you.remember q from return.home and.then
olówaŋ k’uŋ hé káǧe.
song art that he.made.it
‘Do you remember? John came back from Vietnam, then he made the
aforementioned [war] song.’
Waŋží in (53a) is the singular indefinite non-specific article (plural etáŋ); an

alternative translation would be ‘Do you remember any war songs?’, which
is clearly non-referential and avoids the ambiguity of English a(n). In (53b)
waŋ is the singular specific indefinite article (plural eyá), hence the transla-
tion ‘a certain’. In the next sentence, (53c), the song is referred to again and
is marked by kiŋ ‘the’, while in (53d) the other definite article, k’uŋ ‘the
(aforementioned)’ is used. In terms of the immediate common ground, kiŋ
can be used to refer to something in the physical environment which has
not been explicitly mentioned before, as in ‘Please shut the door’, or, as in
(53c) it can refer to a discourse antecedent. It is the default definite article.
The other definite article, k’uŋ ‘the (aforementioned)’, can only be used to
refer to something that has been explicitly mentioned in the discourse; it
could not be used in the ‘Please shut the door’ situation described above, nor
can it be used in associative anaphora, as in ‘I saw Mary’s new house. The
kitchen was huge’. These articles would be linked to the operator projection
of the RP at the RP-level, as shown in Figures 1.14–1.16. In languages without
articles and which use word order to signal the discourse status of an RP,
there is no definiteness operator to be linked to the operator projection.
Rather, information-structural principles will supply the interpretation.
This is analogous to languages without tense which rely on temporal

Table 1.2 Operators in the RP
NucleusR Nominal aspect

CoreR Number, negation
RP Definiteness, deixis
adjuncts to express tense-related notions, but there is no tense operator in

the operator projection of the LSC in them.
The other RP-level operator is deixis, which is primarily instantiated by
demonstratives, and whose canonical function is to locate referents
spatially. This prototypical function relates to the physical environment
part of the immediate common ground, but demonstratives can also refer
to elements in discourse. Languages differ as to whether they allow articles
and demonstratives to co-occur in the same RP. English, for example, does
not allow it (e.g. *that the dog/*the that dog), but Lakhota clearly does, as
shown in (11), (13), (14), (15a), (16) and (53d). Pure definite articles carry
no deictic value, but demonstratives indicate that an RP is minimally
referential and specific and would normally be construed as definite;
hence they carry both operator values when they occur without an
accompanying article.
The operators in the layered structure of the reference phrase (LSRP) may
be summarized as in Table 1.2.
It was mentioned earlier that RPs, unlike PPs and MPs, have a position
somewhat analogous to the positions outside of the core in the LSC (for
English, the PrCS and the PrDP). Its canonical use is to host the possessor RP
in a possessive construction (e.g. John’s book), but it is not limited to posses-
sive constructions. In constructions involving complex nominals, the posses-
sor position may host the equivalent of the core-internal subject of a
proposition, as in Mary’s rejection of the marriage proposal, the rioters’ assault
on the Capitol, a wh-element in the PrCS, as in Whose acceptance of the offer
surprised the faculty the most?, Which assault on the Capitol are you talking about?,
or a frame-setting topic in the PrDP, for example Tomorrow’s march on Wash-
ington will be a very tense affair, Yesterday’s destruction of the town by the earthquake
shocked the nation. This position is called simply the RP-initial position
(RPIP), and it collapses the core-initial subject position, the PrCS and the
PrDP into a single position. That there is only a single position in an RP in
contrast to a sentence is shown by the contrast between Yesterday, who won the
match? vs. *Yesterday’s whose winning of the match. Some left-branching lan-
guages have a corresponding RP-final position (RPFP).
Languages like English do not allow articles or demonstratives to co-occur
with possessors in the RPIP, as shown by *the John’s book/*John’s the book; the
possessor RP signals that the RP is definite, and therefore the genitive RP or
pronoun in the RPIP is attached to both constituent and operator projec-
tions. In order to separate definiteness from possession, it is necessary for
the possessor to occur after the head noun (e.g. the book of John’s vs. a book of

Figure 1.17 General structure of RPs
John’s). In languages like Portuguese, in which articles and possessors co-

occur, the possessor RP would be attached to just the constituent projection.
The derived nominal destruction inherits the argument structure of the
verb destroy from which it is derived (Nunes 1993). If both actor and under-
goer arguments are realized, there are three possible patterns: the destruction
of the village by the earthquake, the earthquake’s destruction of the village, and the
village’s destruction by the earthquake. This illustrates how a nominal can have
core arguments, and it can have locative and temporal adjuncts, for
example the destruction of the village by the earthquake yesterday, the construction
of the bridge in New York City. These adjuncts are in the coreR-level periphery in
the RP, just as in the LSC. There are also RP-level and nuclearR-level periph-
eries, and they are associated with restrictive modification (nuclear) and
non-restrictive modification (RP). The general structure of the LSRP in a
right-branching language is given in Figure 1.17; peripheries are omitted.
Two English RPs are given in Figure 1.18; the first one is a derived nominal
with two semantic arguments and a locative adjunct, while the other
illustrates an RP with modifiers. (‘NML’ stands for ‘numeral’; it is not in an
MP because it does not potentially head a phrase.)
The examples presented thus far are from familiar languages in which RPs
are for the most part headed by nouns or derived nominals. The Tagalog
sentence in (10b) presents a rather different situation; in this case, the
predicator in the nucleus is a noun, while the head of the RP argument
expression is a verb, and there is no nominalizing or verbalizing deriv-
ational morphology. Rather, the occurrence of the verb tumawa ‘laughed’
with the nominative case particle (CSP26) ang induces a type-shift, yielding
the referring expression ‘the one who laughed’. The structure of (10b) is
presented in Figure 1.19.
The declarative IF operator is placed in the immediate post-nuclear pos-
ition, because if it were a yes/no question, the question clitic ba would occur

Figure 1.18 English RPs
Figure 1.19 The structure of (10b) from Tagalog
in second position, that is, Bata ba ang tumawa?’Is the one who laughed a
child?’ (cf. (7)).
In Figures 1.14–1.16 and 1.18 the modifiers expressing properties or char-
acteristics of the head of the RP occur in the nuclear periphery, which is the
locus of restrictive modification in the RP. Non-restrictive modification, on
the other hand, is associated with the RP-level periphery. In simple sen-
tences this is primarily realized through appositive modifiers, for example
Fred Jones, a lawyer, was arrested by the FBI on terrorism charges. In complex
sentences involving relative clauses, restrictive relative clauses occur in the
nuclear periphery and non-restrictive ones in the RP-level periphery. This

correctly predicts that when a complex RP contains a restrictive and a non-

restrictive relative clause, the restrictive clause must be closer to the head of
the RP than the non-restrictive one. This is related to the ordering principle
involving nuclear and clausal operators discussed in Section 1.2.3.1.
(54) a. I called the woman that I met on the train yesterday, who turned out to
be a potential investor, to set up a meeting with our company.
b. *I called the woman, who turned out to be a potential investor, that I met
on the train yesterday to set up a meeting with our company.
1.2.6 Syntactic Templates and the Syntactic Inventory

Syntactic structures must be represented and stored in a grammar. One way
to do this is via phrase structure rules, as in classical transformational
grammar. RRG takes a different approach, representing syntactic structure
as templates and storing them in what it calls the syntactic inventory
(SI). Examples of syntactic templates are given in Figure 1.20.
Syntactic templates are pieces of the LSC which combine to form the
structure of sentences; see Figure 1.21. The templates are language-specific,
but they are composed of the universal and non-universal pieces of the LSC
in Figure 1.20. There are five core templates, three for verbal predicators
and two for non-verbal predicators. The range of core templates in the
syntactic inventory of a language reflects the range of valence patterns of
verbs and other predicators in that language. The most basic ones are given
here: a one-argument template, which would canonically be used with one-
place intransitive verbs (but see below); a two-argument template, which
could be used with a simple transitive verb (two RPs) or with an intransitive
Figure 1.20 Some syntactic templates for English

verb with a second oblique argument such as sit on, arrive at, pray for; a
three-argument template, which can accommodate three RPs (Mary gave
Chris a present), two RPs and a PP (Mary gave a present to Chris) or one RP and
two PPs (Sam ran from the school to the fire station). These possibilities highlight
the advantage of having underspecified templates; if the RPs or PPs were
specified in the core templates, then many more would be needed. Because
English word order is so rigid, especially with respect to the position of the
nucleus, it is possible to include the nucleus in the core template. That
would not be possible in German or Croatian, for example, because the
nucleus can occur in a variety of positions within the clause. The other two
core templates are for non-verbal predicators, which require an auxiliary,
normally be, as discussed in Section 1.2.3.3. The one-place template is for
sentences like The beer is cold or The weather is bad, and the two-place template
is for sentences such as Mary is happy about the election results or Sam is upset
with Sally. Below them are two operator projection templates, the first with
the obligatory operators in English, tense and IF, and the second one adding
an aspectual operator, which could be either progressive or perfect. In the
upper right are the templates for PPs, both predicative and non-predicative,
and there are three RP templates: the first two are for pronouns and proper
nouns, which do not take operators or modifiers, and accordingly lack a
layered structure. The third template is for common count nouns; the
obligatory operators, definiteness and number, are included. Finally, below
them are the most frequent ‘extra-spinal’ positions, the PrDP and the PrCS,
as well as the adjunct and modifier hosting peripheries and the MP. If the
sentence does not have any extra positions (PrDP, PrCS), then the ‘cap’ at
the bottom is employed.
The templates are stored in the SI, and when a semantic representation is
to be mapped to a possible sentence structure, the syntactic representation
is created by assembling the pieces from the SI according to the syntactic
template selection principles in the linking algorithm (see Section 1.6.1). The
construction of the syntactic structure for the sentence in Figure 1.6 is
presented in Figure 1.21.
The arrows indicate how the templates fit together; they do not consti-
tute a ‘derivation’ of the sentence. There is a PrDP, and it contains a PP
with a proper noun. The CLAUSE node in the PrDP template joins with the
CLAUSE node in the PrCS template, and the PrCS hosts a wh-word, what,
which is a type of pronoun. The core node in the PrCS template merges
with the core node of one of the core templates. The semantic representa-
tion of the predicator is the basis for selecting the appropriate core tem-
plate, as will be seen clearly when the linking algorithm is introduced later
in the chapter, and so one would expect a core with two core argument
slots, since eat is a transitive verb. However, one of the arguments of eat is a
wh-expression, and the default situation is for it to appear in the PrCS;
consequently a one-place core is selected. The appropriate operator

Figure 1.21 Combining syntactic templates
projection is selected, and the adjunct temporal PP is assembled and added

to the structure. After insertion of the lexical items, the result is the tree
structure in Figure 1.6. More will be said about template selection in the
description of the semantics-to-syntax linking in simple and complex
sentences in later sections of this chapter, as well as in Chapter 19.
1.2.7 The Layered Structure of the Word

It has been shown that clauses and phrases have layered structures, and the
same is true for the internal structure of words. Cortés-Rodríguez (Chapter 8)
presents a detailed discussion of RRG approaches to morphology, and conse-
quently only two issues will be addressed here, namely, the lexical integrity
hypothesis and the phenomenon of head-marking in the morphosyntax.
One of the first proposals for an RRG approach to morphology was made in
Everett (2002), and an example of the layered structure of the word (LSW)
following his proposal is given in Figure 1.22.
Like phrases (RP, PP, MP) and unlike clauses, the LSW spine has only three
layers: nucleusw , corew, and word. The meaning-bearing root or stem is in
the nucleusw (there may be multiple roots) along with any derivational
morphemes. Inflectional morphology is at the corew-level, and clitics are
attached at the word-level analogous to the Pr/PoDP. In refusals in Figure 1.22,
the verb root refuse and the derivational suffix -al constitute the nucleusw,

Figure 1.22 Layered structure of the word following Everett (2002) and English refusals
and the plural marker is an element in the corew. The difference between
the plural -s in English cats and the possessive clitic -’s in cat’s is reflected in
the LSW: cats would be [WORD [COREw [NUCw /kæt/] {PL}]], while cat’s would be
[WORD [COREw [NUCw /kæt/]][{POSS}]]. As with the LSPP and LSMP, there are no
operators modifying the layers of the word, hence there is no operator
projection in Everett’s model of the LSW. (See Cortés-Rodríguez (Chapter 8)
for a different view.)
1.2.7.1 The Lexical Integrity Hypothesis

The lexical integrity hypothesis (LIH) concerns the ability of the syntax to
have access to the internal structure of words. It is often phrased in terms of
‘can the syntax see into the internal structure of words?’ There is a range of
views on this issue, from the morphology is completely opaque to the syntax to
various degrees of transparency. The completely opaque view seems to be
untenable, because, for example, verb/auxiliary agreement processes require
that the syntax be able to identify the person and number of the subject,
which are coded morphologically. The LSW provides a simple answer to this
question: syntax can see into the corew of the word, but not into the
nucleusw of the word. So from the point of view of the syntax, there is no
difference between refusals and cats, since both are plural count common
nouns. The information that refusals has an internally complex nucleusw and
cats does not is unavailable and plays no role in any syntactic processes. But
the corew-level inflections are accessible. Clitics should also be accessible,
since they are outside of the nucleusw.
1.2.7.2 Head-Marking
The visibility of inflectional morphology to the syntax is perhaps most
clearly observable in head-marking languages. RRG has been concerned with
head-marking grammar since its inception, since Lakhota, one of the
‘founding languages’ of the theory (Section 1.1), is consistently head-
marking. Van Valin (1977) presented an initial attempt at an analysis of
head-marked clause structure, followed by a more detailed account in Van
Valin (1985).27 The basic phenomena are illustrated in (55).

(55) a. Zuzéča kiŋ wičhá-wa-ka-t’iŋ-kte.

snake the 3pl.anim.u -1.a-by.striking-die-pot
‘I will beat the snakes to death.’
a′. Wičhá-wa-ka-t’iŋ-kte.
‘I will beat them to death.’
b. Zuzéča kiŋ Ø-Ø-yaȟtáka-pi.
snake the 3sg.u-3a-bite-pl
‘The snakes bit him/her.’
c. Wičhá-ka-t’a i<bl>útȟiŋ-kte.
3pl.anim.u-by.striking-die <1sg.a>try-pot
‘I will try to beat them to death.’
c′. *Wičhá-wa-ka-t’a i<bl>útȟiŋ-kte.
3pl.anim.u-1SG .A -by.striking-die <1sg.a>try-pot
d. (Zuzéča kiŋ) wičhá-wa-kte-kte.
Snake the 3pl.anim.u-1sg.a-kill-pot
‘I will kill the snakes/them.’
d′. Wičhá-kte i<bl>útȟiŋ-kte.
3pl.anim.u-kill <1sg.a>try-pot
‘I will try to kill them.’
The verb kat’Á ‘beat to death’ (typically with an axe, club, baseball bat, etc.)
is made up of the stative verb t’Á ‘die, be dead’ and the instrumental prefix
ka- ‘by striking’; a more literal translation would be ‘cause to die by
striking’.28 In (55a) there is an object (undergoer) RP zuzéča kiŋ ‘the snake
(s)’, which is cross-referenced on the verb by the third-person plural animate
undergoer prefix wičhá-. The subject (actor) is signalled by the first-person
singular actor prefix wa-; no independent pronoun is required. There is no
case marking on the RP; it has the same form in (55a) as undergoer as it has
in (55b) as actor. The difference in interpretation derives from the different
ways it is cross-referenced on the predicator. In (55a′) the verb constitutes
the entire sentence, and it is with respect to the analysis of examples like
this that theories differ in important ways. Many posit null pronouns
occupying the same positions in the syntax as full RPs which trigger the
agreement/cross-reference on the verb. They are in effect treating these
languages as being dependent-marking with a lot of agreement morphology
(see Van Valin (2013) for detailed discussion). Such an analysis is impossible
in RRG because of the concreteness condition on syntactic representations
introduced in Section 1.1. Rather, as it has since Van Valin (1977), RRG
recognizes the bound affixes on the predicator as being the core arguments
that the syntax is sensitive to. An example of the syntax ‘seeing into the
inflectional morphology of a word’ is the obligatory control construction in
(55c). The first-person singular actor prefix on the linked verb, in this case
wa-, must be omitted, and the actor of iyútȟA ‘try’ must be interpreted as the
actor of the linked verb as well. If both verbs have actor marking, the result,
as in (55c′), is ungrammatical. No RPs or independent pronouns are
required. Only the bound argument markers on the predicator matter for
this construction. Moreover, the construction is oblivious to whether the

Figure 1.23 The structure of wičháwakat’iŋkte ‘I will beat them to death’ in (55a, a′)
predicator is internally complex or not; the mono-morphemic verb kté ‘kill’

is treated exactly the same morphosyntactically as the bi-morphemic
derived verb kat’Á ‘beat to death’, as illustrated in (55d, d′). The structure
of the verb in (55a, a′) is given in Figure 1.23.
If it is the bound argument markers on the predicator which count as the
core arguments for the syntax, a question immediately arises as to the status
of the RPs like zuzéča kiŋ ‘the snake(s)’ in (55). If the valence of the predicator
is saturated by the bound affixes, then there are no valence ‘slots’ available
in the core for full NPs or independent pronouns. If the RP were to be
analysed as being in the core of the clause, then the undergoer argument
in (55a) would be doubly instantiated (RP zuzéča kiŋ ‘the snake(s)’ and wičhá-
‘3planimU’), something that is normally not allowed. A number of solutions
have been offered (see Van Valin (2013) for a critical overview), and RRG has
consistently maintained that the RPs are within the clause but outside of the
core; they are not in clause-external detached positions. Straightforward
evidence for this can be found in wh-questions like (56).
(56) a. Zuzéča kiŋ tuwá Ø-Ø-yaȟtáka-pi he?

snake the who 3sg.u-3a-bite-pl q
‘Who did the snakes bite?’
b. Tuwá zuzéča kiŋ wičhá-Ø-ka-t’a he?
who snake the 3planimu-3sgA-by.striking-die q
‘Who beat the snakes to death?’
c. Hokšíla kiŋ zuzéča kiŋ wičhá-Ø-ka-t’a he?
boy the snake the 3pl.anim.u-3sg.a-by.striking-die q
‘Did the boy beat the snakes to death?’
Lakhota has wh-in situ; there is no displacement of wh-expressions to the

PrCS. This is significant, for two reasons. First, tuwá ‘who’ must be outside
of the core but inside the clause, because it is in the scope of the IF operator
he in either initial or pre-core position. In the polar question in (56c) the
focus of the question can be on the RP hokšíla kiŋ ‘the boy’, which means
that it is clause-internal as well; see the discussion in Section 1.2.3.2
regarding location in a sentence and IF scope. Second, this shows that
Lakhota lacks a PrCS. The positions outside of the core but inside the clause
are termed extra core slots (ECSs) in Van Valin (2013), where they are

Figure 1.24 The structure of (55a)
differentiated from PrCSs in terms of a number of criterial features. The

structure of (55a) is given in Figure 1.24. Having ECSs does not preclude the
possibility of having a PrCS or PoCS as well; for example, Kihara (2017) shows
that Kikuyu, a thoroughly head-marking Bantu language, has both ECSs and
a PrCS.
This analysis raises an interesting question for binding theory, namely, to
what category do the bound argument markers belong? It is assumed in
some generative approaches (e.g. Jelinek 1984; Pensalfini 2004) that they are
pronouns, but this is untenable, because pronouns in an argument position
cannot be bound clause-internally, meaning that RPs and independent
pronouns would have to be in detached positions, which is incompatible
with the facts of polar and wh-questions. It is not necessarily the case that
the bound argument markers are all of the same type. Bresnan andˇ
Mchombo (1987) propose that in the Bantu language Chichewa the object
markers are pronouns, because they cannot be bound clause-internally, but
the subject markers are ambiguous between being agreement when there is
a subject RP present and being a pronoun when there is no subject RP. The
same pattern holds in Kikuyu (Kihara 2017). A similar situation obtains in so-
called ‘pro-drop’ languages like Spanish (Belloro 2007; Kailuweit 2008): the
subject marker counts as agreement when there is a subject RP present,
otherwise as a pronoun. The accusative clitic, however, is pronominal in
standard varieties of Spanish.29

(57) a. (Juan) compr-ó un regalo para María.

buy-3sg.pst a present for
‘Juan/He bought a present for María.’
b. (Juan) lo compr-ó para María, el regalo.
3sg.acc for the present
‘Juan/He bought it for María, the present.’
c. *(Juan) lo compr-ó el regalo para María.
*‘Juan/He bought it the present for María.’
The accusative clitic lo (singular) ~ los (plural) cannot co-occur in the same
clause with an overt object RP; they are in complementary distribution, as in
ˇ
Chichewa and Kikuyu. If both are to appear in a single utterance, the RP
must be in a detached position, either before or after the clause. The subject
marker on the finite verb or auxiliary, on the other hand, can co-occur with
an RP in the core or refer like a pronoun if there is no subject RP. Rather
than analysing these morphemes as sometime being agreement and some-
time being pronominal, Van Valin (2013) argued that they should be con-
sidered to be pronominal anaphors: they can be bound locally by an
argument, like an anaphor, or they can refer independently, like a pronoun,
ˇ
when not bound locally by an argument. In Chichewa, Kikuyu and Spanish,
the subject affixes are pronominal anaphors, while the object markers
are pronouns.
An interesting twist is provided by the dative clitics in Spanish: they are
pronominal anaphors, like the subject markers, and unlike the accusative
clitics, as (58) shows.
(58) a. La música les gusta a los chicos.

the music 3pl.dat like.3sg.pres dat 3pl.m boys
‘The boys like the music.’
b. La música les gusta.
the music 3pl.dat like.3sg.pres
‘They like the music.’
c. *La música gusta a los chicos.
the music like.3sg.pres dat 3pl.m boys
‘The boys like the music.’
d. [CL [PrCS A quién][CORE le gusta la música? ]]
dat who 3sg.dat like the music
‘Who likes the music?’
In (58a) the dative clitic co-occurs in the core with a dative RP, indicating
that it can be bound locally. It can also refer independently, as in (58b), and
(58c) shows that the clitic is obligatory with the verb gustar ‘like’. The wh-
question in (58d) shows that it can also be bound within a simple clause,
which is predicted by the pronominal anaphor analysis.
ˇ
Lakhota is different from Chichewa, Kikuyu and Spanish in that all of the
argument markers, not just the subject markers, are pronominal ana-
phors. They are bound within the clause when there are RPs, and when ˇ
there are none, they refer like pronouns. The contrast between Chichewa

and Kikuyu, on the one hand, and Lakhota, on the other, is significant,
because all three are consistently and thoroughly head-marking, yet they
differ in terms of the binding properties of their argument markers. This is
clearly a parameter of cross-linguistic variation in this type of language.
Another parameter of variation has been proposed in Bohnemeyer et al.
(2016) based on data from Yucatec Maya, another thoroughly head-marking
language. They argue that in Yucatec, RPs are in the core, not in ECSs as in
Lakhota. While they present their conclusions as tentative rather than
definitive, they are nevertheless very intriguing. There are at least two
ways to think about this. On the one hand, it could be taken as evidence
that the bound argument markers do not necessarily saturate the valence
slots in the core completely, as in Figure 1.24, but rather the RP and the
affix together satisfy them. Alternatively, this could be viewed as being
analogous to the variation in the distance between the antecedent (binder)
of a reflexive anaphor and the anaphor itself. English and German permit
only a short distance (at most a core) between the two, for example Nancy
bought a new dress for herself but not *Nancy told Bill to buy a new dress for herself,
whereas this sentence would be perfectly grammatical in Icelandic. With
regard to the status of RPs in the clause, the pronominal anaphors in
ˇ
Spanish, Chichewa, Kikuyu and Yucatec permit the antecedent to be
within the core, whereas Lakhota does not. This constraint on the binding
of pronominal anaphors in Lakhota does not affect reflexive constructions,
because they do not involve any binding; they are purely lexical (Van Valin
and LaPolla 1997, §7.5.1).
Thus, two parameters of typological variation among consistently head-
marking languages have been identified, and this highlights the need for
serious investigation of the typology of head-marking languages, as well as
double-marking languages, which have not been discussed here. (See Van
Valin (2005: 18–19) for some preliminary remarks.) As Nichols commented in
her seminal 1986 paper (116):
I have argued that the theoretical apparatus of classical, traditional,

structural and formal grammar is heavily based on dependent-marked
syntax. If the hypothesis of the universally preferred nature of head-
marked patterns holds true, then we will have to recognize that
describing the world’s languages in standard theoretical terms is not
merely Eurocentric distortion, but in fact forces the unmarked
grammatical structure into a framework devised for the marked type.
1.3 The Structure of Complex Sentences
The term ‘complex sentences’ covers two distinct albeit related phenom-
ena, namely, clause linkage, which involves the combining of
predicator-based units, on the one hand, and complex RPs, which

prototypically includes modification by a relative clause of either the head

of the RP (restrictive) or the RP as a whole (non-restrictive) or
complementation, as in the rumour that Mary stole Bill’s wallet, on the other.
In the analysis of clause linkage, there are three main issues which a
theory must address. First, what are the units that are linked? Second,
what are the possible structural relations between the units in the link-
age? And third, what are the semantic relations that obtain between the
units in the linkage? RRG gives unique answers to all three questions, and
they are presented in Ohori (Chapter 13) and Guerrero (Chapter 14). In
Sections 1.3.1, 13.2 and 1.3.3 just the basics will be presented and several
outstanding issues in the theory of clause linkage will be discussed. The
basic analysis of complex RPs will be the focus of Section 1.3.4, with the
detailed presentation in París (Chapter 15).
1.3.1 Juncture
The answer to the question ‘what are the units of clause linkage’ is
simple and straightforward in RRG: the units of the LSC. This is summarized
in (59).
(59) a. [CORE . . .[NUC . . . ]. . . þ . . .[NUC . . . ] . . .] Nuclear juncture

b. [CLAUSE . . .[CORE . . .]. . . þ . . .[CORE . . .] . . .] Core juncture
c. [SENTENCE . . .[CLAUSE . . .]. . . þ . . .[CLAUSE . . .] . . .] Clausal juncture
d. [TEXT. . .[ SENTENCE. . .]. . . þ . . .[ SENTENCE. . .] . . .] Sentential juncture
Nuclear junctures are made up of multiple nuclei within a single core;

they are complex predicators created syntactically instead of through
derivational morphology. This opposition can be seen clearly in the con-
trast between the Lakhota kat’Á ‘beat to death’(i.e. ‘cause to die by striking’)
in (55), repeated below, and the Mandarin nuclear juncture dǎ sı̌ ‘beat to
death’.
(60) a. Zuzéča kiŋ wičhá-wa-ka-t’iŋ-kte. (¼(55a))

snake the 3pl.anim.u-1sg.a-by.striking-die-pot
‘I will beat the snakes to death.’
b. Wǒ dǎ sı̌ le hěnduō de shé. Mandarin
1sg beat die pfv many prt snake
‘I beat many snakes to death.’
Lakhota kat’Á ‘beat to death’ is the result of a productive derivational

pattern in the morphology of the language and is not a nuclear juncture,
whereas the Mandarin complex predicate is not formed through deriv-
ational morphology but rather consists of two verbs, dǎ ‘beat’ and sı̌ ‘die’
which combine to create ‘cause to die by beating’, the same meaning as
Lakhota kat’Á. Core junctures have multiple cores in a single clause.
(61) a. Hans hat versucht, die Tür aufzumachen. German

has tried the door to.open
‘Hans tried to open the door’.

b. Maria hat Hans überredet, das Auto zu waschen.

has convinced the car to wash
‘Maria convinced Hans to wash the car.’
In (61a) the clause contains two cores, Hans hat versucht and die Tür aufzuma-
chen, while in (b) the two cores are Maria hat Hans überredet and das Auto zu
waschen. Clausal junctures have multiple clauses in a single sentence, as in
(62a), and sentential junctures have multiple sentences in a single discourse
unit, which is labelled here as ‘Text’, as in (62b).
(62) a. I’m aware Sam needs money to buy his wife a new car, but why did he
blow all that money betting on horses?
b. As for Germany, the response to the Covid-19 pandemic has been
surprisingly problematic, and as for the EU as a whole, the situation is grim.
In (62a) there are two clauses in the sentence, each with its own IF
operator; the first clause is an assertion and the second a question. In (b),
each clause has its own frame-setting topic in the PrDP, and that means each
of the conjuncts is a sentence, not just a clause.
It is possible to have all four juncture types in a single utterance, as
in (63).
(63) Poor Sally, she tried to fix the broken picture frame and have it done when
Fred got home, but the glue she tried, it didn’t hold the pieces
together properly.
The first sentence in the sentential juncture contains a clausal juncture, and
the first clause in it contains a core juncture (tried to fix), while the second
clause houses a nuclear juncture (have it done), and a temporal adverbial
clause (when Fred got home) modifying the core containing the nuclear junc-
ture. The second sentence contains a nuclear juncture (hold the pieces
together), which, like the one in the first sentence, is an example of an
English resultative construction, like wipe the table clean, paint the fence white.
A rough approximation of the constituent projection of (63) is given in
Figure 1.25. Details of phrases are not spelled out, and the conjunctions
are not represented.
Figure 1.25 The constituent projection for the multi-juncture utterance in (63)

1.3.2 Nexus
The structural relations between units in clause linkage are termed nexus
relations in RRG. Traditionally, two nexus relations are recognized,
namely, coordination and subordination. With respect to clauses, coordin-
ation means that the two clauses are independent of each other and each
can stand on its own as an independent utterance. It is a relation between
two wholes. Subordination, on the other hand, is a part–whole relationship,
in which the subordinate clause functions either as an argument of the
matrix clause or a modifier of it in some way; subordinate clauses cannot
stand on their own as independent utterances.
RRG has introduced two innovations with respect to nexus relations. First,
it applies them to the four layers of the clause: nucleus, core, clause and
sentence. Hence there is a contrast between the two traditional nexus types
at each of the four levels. Second, it posits a third nexus type,
cosubordination, which is a kind of dependent coordination.30 The
dependence is operator sharing: the linked unit must share an operator at
the level of juncture with the licensing unit. For English clauses, the relevant
operators are tense and IF. It is crucial, first and foremost, to distinguish
operator sharing from operator identity. This is illustrated in (64) and (65).
(64) a. Bill bought a new car yesterday, and he sold his old one, too.
b. Bill bought a new car yesterday, and did he sell his old one, too?
c. Bill bought a new car yesterday, and he will sell his old one tomorrow.
d. Bill bought a new car yesterday, and will he sell his old one tomorrow?
e. Did Bill buy a new car yesterday and will he sell his old one tomorrow?
(65) a. Bill bought a new car yesterday and sold his old one, too.
b. Did Bill buy a new car yesterday and sell his old one, too?
b′. *Did Bill buy a new car yesterday and sold his old one, too?
c. Bill bought a new car yesterday and will sell his old one tomorrow./*?
d. *Did Bill buy a new car yesterday and will sell his old one tomorrow?
In (64a) both clauses are past tense, and both have declarative IF; this is a
case of operator identity, not operator sharing. The two clausal operators
happen to be the same, but they are not obligatorily the same. This is shown
in (64b), where the tense is the same in both but IF differs, in (64c) where
tense differs but IF is the same, and in (64d) where both are different in each
clause. Both clauses can be questions, as in (64e), but the interrogative IF
must be indicated separately in each clause. There are no dependencies,
operator or otherwise, between the two clauses. Coreference between Bill
and he is not obligatory, despite being strongly preferred; in a context in
which Sam is the main discourse topic and the issue of car selling and
buying, he could be interpreted as referring to Sam rather than Bill.
This contrasts sharply with the examples in (65), which differ from those
in (64) in that the ‘subject’ of the second clause has been omitted under
identity with the ‘subject’ of the first clause.31 The sentences in (64a) and
(65a) look superficially similar, since they both have the same tense and IF in

both clauses, but they are not the same. In (65) the IF must be the same in
both clauses: in (65b) the tense operator is at the beginning of the clause
signalling interrogative IF, and the verbs in both clauses are bare infinitives,
which means they obligatorily share tense and IF in this interrogative
utterance. Sharing tense is not obligatory with declarative IF, as (65c) shows,
but the second clause cannot be interpreted as a question (hence ‘*?’), unlike
(64d). The examples in (65b′), (c) and (d) are all ungrammatical, because the
two clauses differ in IF, and this contrasts with (64b) and (d). Thus, the
sentences in (64a) and (65a) are both clausal junctures, but they differ in
nexus type: (64a) is coordination, while (65a) is cosubordination, because the
linked clause must have the same IF as the licensing clause and is therefore
dependent on it.
There is another important contrast illustrated here. Both (64) and (65) are
instances of conjunction, which is a kind of grammatical construction. Coord-
ination and cosubordination, as well as subordination, are abstract linkage
relations which are instantiated by a range of formal construction types. The
Mandarin construction in (60b) is an example of cosubordination at the
nuclear level; the two verbs dǎ ‘beat’ and sı̌ ‘die’ are followed by the perfective
aspect marker le, which obligatorily has scope over both of them. This is an
instance of an obligatorily shared operator at the level of juncture (perfective
aspect is a nuclear operator), and consequently the nexus type is cosubordina-
tion. The formal construction type is serial verbs, not conjunction. There is no
one-to-one correspondence between juncture and nexus combinations and
formal construction types. A useful analogy is to traditional grammatical
relations: they are abstract relations which are instantiated formally in a
variety of ways, namely, word order, case marking, cross-reference (as in
head-marking languages), agreement (as in ‘pro-drop’ languages) and even
tone (e.g. Maasai), but ‘subject’ is not necessarily identified or correlated with
a particular case or agreement or a position in the clause. The same is true of
clause-linkage relations. Coordination and cosubordination are not always
realized via conjunction, just like subordination is not always associated with
embedding, as will be shown later in this discussion.
This brings up a fundamental difference between the RRG theory of
complex sentences and other approaches. The following equivalence is
widely assumed: dependent ¼ embedded ¼ subordinate. Terms like ‘depend-
ent clause’, ‘embedded clause’ and ‘subordinate clause’ are used as syno-
nyms. In the RRG system, on the other hand, ‘dependent’, ‘embedded’ and
‘subordinate’ are not synonyms. Neither are ‘conjunction’ and ‘coordin-
ation’, despite common usage of them as synonyms. Coordination may be
realized in constructions which do not involve conjunction, as will be seen
in the discussion of core-level negation and at the end of this section.
Crucially, a dependent unit is not necessarily embedded or subordinate.
This can be seen clearly in (65); the second clause is dependent upon the
first clause for its IF value; if the first clause is an assertion, as in (65a, c),
then the second clause must be interpreted as an assertion as well, and if the

first clause is a question, as in (65b), then the second must be a question, too.
This is a particularly telling example, because the verb in the second clause
is a bare infinitive and accordingly the second clause is dependent on the
first clause for its IF value but also for its tense. If the tense in the first clause
changes, as in Will Bill buy a new car tomorrow and sell his old one, too?, the tense
interpretation of the second clause necessarily changes, too. Thus, the
second clause in (65) is dependent on the first clause, but it is neither
embedded nor subordinate.
The RRG treatment of sub-clausal junctures, specifically core junctures, is
also very different from that of other approaches. These include what are
known as ‘obligatory control’ and ‘raising’ or ‘exceptional case-marking’
constructions. These core junctures have an important property: they
require a shared core argument. This is illustrated in (66).
(66) a. Sam tried to fix the car. Shared argument: Sam ¼ actor of

try, fix
a′. *Sam tried (for) Bill to fix the car.
b. Sam persuaded Bill to fix the car. Shared argument: Bill ¼ undergoer
of persuade, actor of fix
b′. *Sam persuaded Bill (for) Tom to fix the car.
(67) a. Sam seems to have fixed the car. Shared argument: Sam ¼ core
argument of seem, actor of fix
a′. *It seems (for) Sam to have fixed the car.
b. Bill believes Tom to have fixed the car. Shared argument: Tom ¼ core
argument of believe, actor of
fix
b′. * Bill believes Tom (for) Sam to have fixed the car.
There are two types of argument sharing in non-subordinate core junc-

tures: (66) illustrates a controller–pivot relationship, whereby the controller
in the licensing core, Sam in (66a) and Bill in (66b), supplies the interpret-
ation for the missing argument in the linked core. The ungrammatical
examples lack a shared argument in the linked core. In (67) the nature of
the sharing is different, in that a semantic argument of the predicator in the
linked core is realized as a syntactic core argument in the licensing core. The
ungrammatical examples lack a shared argument. Subordinate core
junctures may but need not have a shared argument, whereas peripheral
ad-core subordinate cores require one.
(68) a. Susan regretted insulting Donald. Core subordination

b. Susan regretted the teacher’s insulting Donald.
(69) a. Chris was stopped after running a red light. Peripheral ad-core
subordination
a′. *Chris was stopped after his/Mary’s running a red light.
b. Sally brushed her teeth before meeting Max for dinner.
b′. *Sally brushed her teeth before her/Sam’s meeting Max for dinner.

Argument sharing will be discussed in more detail in section 1.7.2, on

linking syntax and semantics in complex sentences, as well as in Guerrero
(Chapter 14).
There are two more ways in which the RRG analysis of (66) and (67) departs
from the mainstream. First, the structures in (68) and (69) are examples of
core subordination, and the standard analysis of (66) and (67) is that they
involve subordination as well. The subordinate gerund in (68) can be passi-
vized and it-clefted.32
(70) a. Insulting Donald was regretted by Susan.

a′. The teacher’s insulting Donald was regretted by Susan.
b. It was insulting Donald that Susan regretted.
b′. It was the teacher’s insulting Donald that Susan regretted.
The infinitive in (66a) cannot.
(71) a. *To fix the car was tried by Sam.

b. *It was to fix the car that Sam tried.
The passive test is not applicable to the structures in (66b) and (67), but the it-
cleft test is applicable; they all fail.
(72) a. *It was to fix the car that Sam persuaded Bill.

b. *It is to have fixed the car that Sam seems.
c. *It is to have fixed the car that Bill believes Tom.
This appears to be clear evidence that the infinitives in these constructions

are not subordinate but rather they are non-subordinate core junctures,
contra the mainstream analysis. Interesting evidence in support of this
conclusion comes from try, which can license more than one type of linked
core. It can also take a gerund as a linked core, and like the gerund with
regret, it passes both tests.
(73) a. Sam tried fixing the car.

b. Fixing the car was tried by Sam.
c. It was fixing the car that Sam tried.
There is thus a clear contrast between try þ infinitive vs. try þ gerund with
respect to nexus. This illustrates an important point: a given predicator can
license more than one juncture–nexus combination (see Ohori, this volume,
Chapter 13, Van Valin 2005: 210).
The second difference also concerns the nexus type of these construc-
tions. The difference between the two non-subordinate nexus types is: if
there is a core operator (e.g. deontic modality) in the clause, it must have
scope over all of the cores in cosubordination but not in coordination.
This only applies when the primary participant (‘subject’) is the shared
argument between the cores, as in (74a–c); it does not apply to construc-
tions where the primary argument is not the shared argument, as in
(74d–e).

(74) a. Sam must [is obliged to] try to fix the car.

b. Sam can [is able to] try to fix the car.
c. Sam must [is obliged to] want to be able to walk again, before he will be
able to do so.
d. Sam must [is obliged to] persuade Bill to fix the car.
e. Sam can [is able to] persuade Bill to fix the car.
In (74a, b) Sam is the shared argument, and Sam’s obligation or ability is to

try to fix the car, not just try something unspecified, and accordingly the
deontic modal has scope over both cores. In (74c) Sam is again the shared
argument, but in this sentence the scope of the deontic modal operator in
the first core does not extend to the second core, which is shown by the fact
that it has a different deontic operator. In (74d, e) his obligation or ability
is to persuade Bill to fix the car, but the modals do not affect the relation-
ship between Bill and fixing the car; in other words, (74d) doesn’t mean
‘Sam is obliged to persuade Bill to be obliged to fix the car’, nor does (74e)
mean ‘Sam is able to persuade Bill to be able to fix the car’. The modals in
the first core do not have scope over the second core, and in fact it is
possible to have distinct deontic operators in each core (i.e. Sam can [is able
to] persuade Bill to be obliged to fix the car). Thus, there is obligatory operator
dependence at the level of juncture in (74a, b) but not in (74c–e), and
therefore the former are instances of cosubordination and the latter
coordination.
This conclusion is supported by the scope of core-level negation, the
default interpretation of negation. The issue is whether negation in the first
core can license a negative polarity item (NPI) in the second core; NPIs
include any (Bill didn’t buy any flowers vs. *Bill bought any flowers) and a red cent
(Bill didn’t spend a red cent vs. *Bill spent a red cent).
(75) a. Bill didn’t try to buy any/some flowers. (any > some)
b. Bill didn’t try to spend a red cent.
c. Bill didn’t persuade Sam to buy any/some flowers. (some > any)
d. ?Bill didn’t persuade Sam to spend a red cent.
(‘x > y’ means ‘x is preferred over y’.) In the cosubordinate examples, (75a, b),
negation in the first core can clearly license an NPI in the second. On the
other hand, in (75c, d) the licensing of the NPI in the second core is less
acceptable than in the first two examples. Core-level negation is more easily
shared across the cores in the cosubordinate rather than the coordinate core
juncture. This is a case of coordination not being realized by conjunction.
1.3.3 Some Issues Concerning Juncture and Nexus

Coordination and subordination are found at all four levels of juncture, as
noted at the beginning of the previous section. Cosubordination is not
possible at the sentence level, because there are no sentence-level operators,
and accordingly obligatory operator sharing is ruled out. Hence there are

Table 1.3 Nexus types
þembedded (in a core, clause or periphery) Subordination

‒embedded, þdependent (obligatory operator sharing at the level of juncture) Cosubordination
‒embedded, ‒dependent (no operator sharing at the level of juncture) Coordination
Figure 1.26 The structure of (64e): English clausal coordination
only eleven juncture–nexus combinations instead of twelve. These will be

discussed more in Section 1.3.4.
The three nexus types may be characterized as shown in Table 1.3.
1.3.3.1 Representing Juncture–Nexus Types

Since cosubordination is distinguished from coordination by obligatory
operator sharing at the level of juncture, the operator projection is very
important for the representation of complex sentences. The constituent and
operator projections are more or less mirror images of each other. The only
difference is that when there are multiple operators at a given level, each of
those operators is represented by a distinct node. In the examples in
Figures 1.26 and 1.27, there are two clausal operators, tense and IF, and so
there are two clause nodes in the operator projections of these sentences.
Coordination and cosubordination differ in terms of operator sharing, and
this is represented by having a superordinate node at the level of linkage to
which the operator applies and thereby has scope over the units. This can be

Figure 1.27 The structure of (65b): English clausal cosubordination with two shared operators
seen clearly in Figure 1.27, in which the superordinate clause nodes indicate
that tense and IF have scope over both clauses. There is no such superordin-
ate node in Figure 1.26, because each clause has its own tense and IF
operators. The structure of (65c) is interesting in that each clause has its
own tense but shared IF indicated by the position of the tense operator in
the first clause; it is given in Figure 1.28.
At sub-clausal levels the same patterns hold. The structure of (60b), a case
of nuclear cosubordination in Mandarin, is given in Figure 1.29.
In sub-clausal junctures, all higher layer operators are necessarily shared;
in a nuclear juncture like this one, all core and clausal operators have scope
over the nuclei, and similarly, in a core juncture all clausal operators have
scope over the cores.
1.3.3.2 Symmetry in Clause Linkage

The picture of clause linkage given in (59) is not the whole story. The
combinations of nucleus þ nucleus, core þ core, etc., reflect the flat struc-
tures of coordination and cosubordination. These two nexus types are inher-
ently symmetrical: only units of the same type can be linked in coordinate
and cosubordinate relations. The same is not true of subordination, how-
ever. Complementation is perhaps the prototypical case of subordination,

Figure 1.28 The structure of (65c): English clausal cosubordination with one shared operator
Figure 1.29 The structure of (60b): Mandarin nuclear cosubordination

that is, the use of a clause as a core argument, and it is asymmetrical: a

larger unit is embedded in a smaller unit.
(76) a. [CORE Larry believes [CLAUSE that Sam doesn’t like ice cream]]
b. [CORE [CLAUSE That Sam doesn’t like ice cream] surprised his friends]
In a tree diagram the embedded clause would be a daughter of the core

node, and so this type of subordination will be referred to as daughter
subordination. These examples also illustrate a point around which there
has been some confusion. The issue is, when there is asymmetric linkage, is
the linkage type defined in terms of the size of the linked unit or the size of
the licensing unit? In (76), the linked unit is a clause, and the licensing unit
is a core. The clauses function as core arguments of the predicator in the
nucleus, and therefore these are core junctures; in asymmetrical linkages
the juncture type is determined by the licensing unit. Otherwise the analysis
is very inconsistent. Symmetrical core subordination is illustrated in (77).
(77) a. [CORE Larry regrets [CORE offending Max]]

b. [CORE [CORE Offending Max] is the least of Larry’s worries.]
The linked unit in these examples is a gerund, which is a nominalized core,

and accordingly since a core is embedded in a core and functions as a core
argument, this is clearly a core juncture. If one analyses (76) as a core
juncture, then (76) and (77) are parallel constructions: both have a phrasal
core argument larger than a simple RP, and despite the difference in the size
of the linked unit, they are both core junctures. If, on the other hand, (76)
were to be analysed as a clausal juncture, then (76) and (77) would not be
parallel constructions, which is unsatisfactory.
Further support for the analysis of (76) and (77) as core junctures comes
from the variants in (78), which are symmetrical like (77).
(78) a. [CLAUSE [CORE Larry believed] [PERIPHERY yesterday] [CLAUSE that Sam doesn’t
like ice cream CORE] CLAUSE]
b. [CLAUSE [CORE It surprised his friends] [CLAUSE that Sam doesn’t like ice cream]
CLAUSE]
These are extraposition constructions, and one way to look at them is as a

way to resolve asymmetrical linkage as symmetrical. In (76a) the comple-
ment clause appears to be within the core, but the structure is actually
ambiguous, because the embedded clause can either be inside the core, as
represented in (76a) or it could be outside of the core and a direct daughter
of the higher clause node. There is evidence supporting the position of the
embedded clause being outside of the core. It is given in (78a). English
normally restricts adjuncts in the core-level periphery to appear after all of
the core elements, and accordingly the fact that the temporal adverb yester-
day is strongly preferred before the embedded clause (cf. ??Larry believed that
Sam doesn’t like ice cream yesterday) indicates that it is outside of the core.
Because RRG is a monostratal theory with no derivations or movement rules,

the option of generating the clause core-internally and then moving it to an

extraposed position after the adjuncts in the core-level periphery is not
available. The appropriate option in RRG is to link the embedded comple-
ment clause to the position outside of the core, which then accommodates
the placements of the adjuncts. This also results in a symmetrical linkage.
There is no ambiguity with respect to the core-internal position of the
embedded clause in (76b), which, while fully grammatical, is unlikely to
occur in conversation. Rather the preferred form is (78b), in which
the embedded clause is located outside of the core, yielding a
symmetrical linkage.
The other type of subordination is adverbial subordination (see Guerrero,
Chapter 14, for detailed discussion). In RRG the general term ‘adverbial’ can
be replaced by ad-nuclear, ad-core or ad-clausal subordination, and
the cover term for these is peripheral (ad-)subordination, since the
clauses occur in the periphery associated with the unit being modified.
The question of symmetry of linkage is not as clear-cut with peripheral
subordination; the core- and clause-level peripheries can house adverbials
as in (78a), PPs as in (34)–(36), and ad-subordinate clauses, as in (63). The ad-
subordinate clauses are outside of the core or clause, depending on the level
they modify, and there do not appear to be any issues in relation to sym-
metry. As noted in Section 1.2.4.1, there are ordering constraints on them,
such that core-level adjuncts occur closer to the core than clause-level
adjuncts (see (34)–(36), Figure 1.11), and the same holds true for ad-core
and ad-clausal subordinate clauses.
(79) a. Susan was upset with Paul for not washing the dishes after he got home
from the gym even though he had promised to do so.
b. Susan was upset with Paul for not washing the dishes even though he had
promised to do so after he got home from the gym.
There are two ad-subordinate clauses in these sentences: after he got home
from the gym, a temporal ad-core subordinate clause, and even though he had
promised to do so, a concessive ad-clausal subordinate clause. The two sen-
tences don’t have the same meaning: in (79a) the ad-core clause modifies the
main clause, so that the dishwashing was supposed to happen after the gym
due to a prior promise, whereas in (79b) the promise was made after the gym
and the time of the dishwashing is unspecified. The clause-level concessive
adjunct blocks the core-level temporal adjunct from modifying the main
clause, and it must be interpreted as modifying the core of the ad-clausal
clause, hence the difference in meaning.
It appears there are strong restrictions on what can modify the nucleus;
no nuclear-level PPs or ad-subordinate clauses modifying the nucleus have
thus far been encountered. Adjuncts in the nuclear-level periphery appear to
be restricted to adverbials.
There are some quite extreme instances of asymmetrical embedding; two
examples were given in (12), (12′) and Figure 1.15, in which whole clauses

and sentences are used as attributive modifiers in the MP, filling a slot that
would normally be filled by an adjectival modifier, possibly with a degree
modifier. A radically asymmetrical linkage can be seen in what Everett and
Kern (1997) and D. Everett (2008) term ‘intentional state constructions’ in
Wari’, the last viable Chapakuran language spoken in the Amazon in Brazil
and Bolivia. A simple Wari’ sentence is given in (80a) and an example of an
intentional state construction is given in (80b), taken from D. Everett (2008).
Basic word order is V-CL OS, and stress is on the last syllable of each word
(indicated by italics).
(80) a. Mi′ noni -onj con hwam hwijima′i mon tarama’i.

give 3pl.s.rp/p-3pl.o.m prep.3sg.m fish children coll man
‘The men gave the children fish.’
b. [Ma′i co mao nai -inj Guajaráj]
that.prox.hearer m/f.rp/p go(sg) 3sg.rp/p-3n Guajará (Brazilian city)
nak -naml ′oro narima′l taramaxiconk.
3sg.rp/p-3pl.f coll woman chief
‘“Who went to Guajará?” (said) the chief to the women.’
In (80a) the sentence begins with a verb followed by a clitic cluster cross-
referencing the direct core arguments, which is followed by an oblique core
argument followed by the two direct core arguments. In the intentional
state construction in (80b) the sentence begins with the wh-question ‘who
went to Guajará?’ followed by the clitic cluster and the two direct core
arguments. What is striking about (80b) is that the main clause has no verb
or predicate in it, and the slot for the nucleus is filled by the clause
representing the content of the speech. Everett presents a number of argu-
ments in favour of this analysis and shows that alternative analyses all fail
to capture the Wari’ facts. A simplified representation of (80b) is given in
Figure 1.30.
Here a clause is embedded in a nucleus, an extreme violation of the
tendency toward symmetrical embedding, but it is allowed by the theory
Figure 1.30 Proposed structure for (80b) in Wari’ (Everett 2008)

as a highly marked construction. The RRG analysis that Everett gives

accounts in a principled way for all of the highly unusual properties of
this construction.
1.3.4 The Syntax–Semantic Interface in Clause Linkage

It was mentioned at the beginning of the previous section that there are
eleven juncture–nexus combinations. They may be organized into a hier-
archy in terms of the tightness of the relation between the units; this is
known as the Interclausal Syntactic Relations Hierarchy, and it will be
discussed in Section 1.3.4.1. The semantic relations holding between the
units in the linkage can also be organized into a hierarchy, yielding the
Interclausal Semantic Relations Hierarchy, which is the topic of Section
1.3.4.2. The two will be combined to create the Interclausal Relations Hier-
archy in Section 1.3.4.3. (See also Ohori, Chapter 13.)
1.3.4.1 The Interclausal Syntactic Relations Hierarchy

Ranking the juncture–nexus combinations in terms of how tightly bound
the units are to each other in the linkage yields the Interclausal Syntactic
Relations Hierarchy (ISynRH). One aspect of it is straightforward: in terms
of the levels of juncture, nuclear junctures are tighter than core junc-
tures, core junctures are tighter than clausal junctures, and clausal
junctures are tighter than sentential junctures. With respect to nexus
types, coordination is the ‘loosest’ because there can be no dependencies
at the level of juncture between or among the units linked. That leaves
the two types of subordination, daughter and peripheral, and cosubordi-
nation. The clearest comparison is at the clause level, because clauses
have more freedom of placement than cores and nuclei, and since there is
no cosubordination at the sentence level, it is irrelevant to this compari-
son. Accordingly, at the sentence level, coordination is looser than
subordination.
Of the three nexus types peripheral ad-subordinate clauses have the
greatest freedom in terms of where they can occur. It has been suggested
that a universal property of these clauses is that they can occur either before
or after the licensing (‘main’) clause; this is not the case with daughter
subordinate clauses, which in some languages have a fixed position at one
edge of the clause or the other, depending on the direction of branching,
and in others can appear in extraposed portions, as exemplified in (76) and
(78). The core- and clause-level peripheries are less rigidly organized than the
core or clause, and their constituents are more loosely related to each other
and to the core or clause they modify. Peripheral ad-subordinate clauses
need not share any operators with the modified clause; as far as IF is
concerned, they are normally presupposed therefore are not the focus of
assertions or questions (see Bentley (Chapter 11)). They are optional modi-
fiers rather than semantic arguments of the predicator in the nucleus like

daughter subordinate clauses. Hence it is reasonable to conclude that per-

ipheral ad-subordinate clauses are less tightly linked to the other unit in the
juncture than daughter subordinate clauses.
Peripheral ad-subordinate clauses are clearly more loosely linked than
cosubordinate clauses, which have little or no possibility of variable loca-
tions in the sentence. They are flat structures which must be dependent on
the licensing unit for an operator at the level of juncture. Neither peripheral
nor daughter subordinate clauses are required to share any operators with
the licensing clause. Thus it appears that the ranking of tightness of the
nexus types is coordination (least) > peripheral ad-subordination > daugh-
ter subordination > cosubordination (most).
This ranking, together with the one for juncture levels, generates an inter-
esting prediction, first made in Foley and Van Valin (1984), namely, there is a
default or unmarked nexus type for each level of juncture, such that if a
language has only one nexus type at a given level of juncture, then it will be
the unmarked one. So if a language has only one type of nexus at the clause or
sentence level, it will be coordination. If a language has only one nexus type at
the core level, it will be subordination, and the question then arises, which
kind? The prototypical subordinate clause is a complement clause which is a
semantic argument of the predicator in the nucleus. However, it has been
claimed that there are languages that lack this kind of subordination (e.g. Hale
(1976) for Warlpiri, D. Everett (2005) for Piraha)~ but do have ‘adjoined’ struc-
tures, which would be analogous to peripheral ad-subordination. Finally, at

the nuclear level, the unmarked nexus type would be cosubordination, and it
seems very often to be the case that a language has only one type of nuclear
juncture, which reveals itself to be cosubordination.
The ISynRH is given in Figure 1.31. It is important to emphasize that
languages vary in terms of which juncture–nexus types they exhibit; most
languages do not have all of them, and which ones a given language has is
an empirical question.
Above nuclear cosubordination in terms of tightness is derivational
morphology, which is outside of the domain of complex sentences. Thus,
the Lakhota example in (55) and the Mandarin example in (60b), both of
which mean ‘beat to death’, contrast structurally in that the Lakhota form
kat’Á is a morphological construction whereas Mandarin dǎ sı̌ is a syntactic
construction, an instance of nuclear cosubordination (compare Figures 1.24
and 1.29).
1.3.4.2 The Interclausal Semantic Relations Hierarchy

In Section 1.3.2 an analogy was made between juncture–nexus types and
traditional grammatical relations based on the shared property that they are
abstract relations that can be realized via a variety of formal means. Another
way they are similar is that they are both syntactic in nature but serve to
convey semantic information: grammatical relations code semantic roles
like agent and patient, and similarly juncture–nexus types express semantic

Figure 1.31 Interclausal Syntactic Relations Hierarchy
relations like causality, purposive, jussive, etc. These relations can be

thought of as the ‘thematic relations’ of clause linkage. The following is a
revision of the interclausal semantic relations in Van Valin (2005, §6.6), and
they are organized into clusters based on shared semantic properties.
(81) Interclausal semantic relations

a. Single actions
1. Causative [1]: one state of affairs makes another state of affairs
happen, and the two are subparts of a single state of affairs, e.g. (60b),
Mary wiped the table clean.
2. Modifying sub-actions
i. Method: the way an action is carried out, e.g. Mary wiped the table
clean, Tom sliced the box open.
ii. Manner: the manner in which a motion event is carried out, e.g.
Bill entered the house limping.
3. Phase: a separate verb describes a facet of the temporal envelope of a
state of affairs, specifically its onset, its termination, or its continuation,
e.g. Sam started singing, Nancy kept crying, Oliver finished eating dinner.
b. Multiple actions
4. Simultaneous
i. Non-motion, e.g. I can eat dinner watching TV, Marie falls asleep
listening to music.

ii. Motion: motion accompanying another action, e.g. Lakhota yuhá

‘have’ þ hí ‘arrive coming’ ¼ ‘come while holding/possessing
something’
iii. Position: stance while doing an action, e.g. Sally stood singing, Bill
sat reading the newspaper.
5. Sequential, e.g. Close and lock the door, Sally will prepare and cook the fish.
6. Causative [2]: one state of affairs causes another state of affairs to
happen, and the causing state of affairs is distinct but unspecified
from the caused state of affairs.
i. Unmediated: Mary made Sam wash the dishes.
ii. Voluntary: Mary had Sam wash the dishes.
iii.Facilitating: Mary let Sam wash the dishes.
iv. Assisting: Mary helped Sam wash the dishes.
c. Endeavour
7. Attempt, e.g. Fred tried/attempted/endeavoured to make the cake without a mix.
8. Success, e.g. Fred managed/happened to make the cake without a mix, Fred
succeeded/lucked out in making the cake without a mix.
9. Failure, e.g. Fred failed to make the cake without a mix.
d. Intentions
10. Refusal, e.g. Fred refused/declined to make the cake without a mix.
11. Psych-action: a mental disposition regarding a possible action on the
part of a participant in a state of affairs, e.g. Mary decided to take the job
offer, Sally forgot to close the window, Andy wants to be examined by a
specialist.
12. Purposive: one action is one with the intention of realizing another
state of affairs, e.g. Harry went to the store to buy some beer, Susan brought
her own wine to drink.
e. Bringing about
13. Causative [3]: the bringing about of a state of affairs by a distinct state
of affairs, e.g. Maria taught Fred to speak Brazilian Portuguese, Marilyn’s
arrogant attitude caused Sam to leave the reception early.
14. Jussive: a verbal causative, i.e. the expression of a command, request
or demand, e.g. The teacher asked the student to stop talking, The general
ordered the troop to attack the enemy.
15. Permissive: making a state of affairs possible by removing a barrier to
it or directly licensing it, e.g. The rancher opened to gate to allow the cattle
to move to a new feedlot, The city council permitted the demonstrators to
gather in the park for their demonstration.
16. Injunctive: making a state of affairs impossible by putting up a
barrier to it or directly enjoining it, e.g. The security fence prevented the
rioters from reaching the Capitol, The fireman stopped Sam from going back
into the burning house.
f. Perception
17. Direct: the unmediated apprehension of some action, event or situation
through the senses, e.g. The security guard saw Tom leave work early.
18. Indirect: the deduction of some action, event or situation from
evidence of it, e.g. (looking at an empty executive parking place) I see
that Tom left work early today.

g. Intentionality
19. Propositional attitude: the expression of a participant’s attitude,
judgement or opinion regarding a state of affairs, e.g. Many people
believe that politicians are dishonest, Max considers Don to be a loser, Sports
fans want very much for their team to win.
20. Cognition: an expression of knowledge or mental activity, e.g. Bill
knows that he is in the wrong this time, Sally is thinking about how to fix the
problem with the microwave.
21. Emotion: an expression of the content of an emotional state, e.g. Sam
is happy/loves it that the pandemic is finally over. Larry is afraid/fears that the
corona virus will return.
h. Speech
22. Indirect discourse: an expression of reported speech, e.g. Frank said
that his friends are corrupt.
23. Direct discourse: The direct quotation of a speech event, e.g. Frank
said, ‘My friends are corrupt.’
i. Locational
24. Space: the spatial location of a state of affairs, e.g. Matilda lived her
whole life near where she was born.
25. Time: the temporal parameters of a state of affairs, e.g. Sally met Bob for
a drink after she left the office.
j. Circumstances
26. Reason: the motivation or cause for a state of affairs, e.g. The baby cried,
because it was hungry.
27. Conditional: an expression of what consequence would hold, given a
particular state of affairs, e.g. If it rains, we’ll have to stay home.
28. Concessive: the state of affairs specified in the main clause holds
unexpectedly, given the state of affairs specified in the subordinate
clause, e.g. Tim made it to work, even though it was snowing heavily.
k. Temporality
29. Simultaneous: one state of affairs is temporally coterminous with
another, e.g. Lisa yawned loudly while Dan was talking.
30. Sequential: one state of affairs follows another temporally, with or
without any temporal overlap, e.g. Marge finished breakfast, and then the
repairman was at the door.
31. Unordered: the temporal relation between states of affairs is
unspecified, e.g. Sam watched a football match, and Sally chatted with
her mother.
There are a number of shared semantic properties reflected in this list.

First, for the most part, groups (a) through (h) are either verb-governed (e.g.
(c), (e), (f ), (g) and (h)) or there are restrictions on one or both of the
predicators involved (e.g. (a) and (d)). This contrasts sharply with (i) through
(k), which do not depend upon the content of the clauses being linked.
Second, the notions expressed in (i) and (j) are the same as ones expressed
by the core-level and clause-level peripheral adjuncts discussed in Section
1.2.4.1. Third, the relations in (j) and (k) are essentially discourse relations of
the kind posited in Rhetorical Structure Theory (Mann and Thompson 1988),

and thus the hierarchy constitutes a link between clause-level syntax–

semantics and discourse relations (see also Winther-Nielsen 1995, 2021).
Fourth, groups (a)–(c) have the Macro-Event Property (MEP) (Bohnemeyer
and Van Valin 2017), while (d) through (k) lack it.
The order of the relations in (81) is not random but rather reflects how
close or loose the semantic relation between the unit is. In group (a) the
units in the linkage are taken to be subparts of a single action, whereas all
other groups involve multiple actions. One piece of evidence in favour of
ranking the relations in group (a) at the top of the list is the fact that these
relations are lexicalized in many cases, or expressed by derivational
morphology, for instance Lakhota instrumental prefixes express a combin-
ation of causative [1] and method. These interclausal semantic relations
can be represented in a hierarchy with the relations expressing the closest
semantic ties at the top and those indicating the loosest relationship at
the bottom.
This is a complex hierarchy, and it has been argued that it is motivated by
a number of more basic hierarchies, which are discussed by Guerrero in
Chapter 14.
1.3.4.3 The Interclausal Relations Hierarchy

It has been observed, since at least Silverstein (1976), that there is a funda-
mentally iconic relation governing the interaction of syntax and semantics
in clause linkage. Specifically, the closer the semantic relationship is
between the units in the linkage, the tighter the morphosyntactic integra-
tion of the units will be. Thus, closeness of meaning correlates with the
extent of the morphosyntactic fusion of the units. This is captured in the
Interclausal Relations Hierarchy (IRH) in Figure 1.32, which is composed of
the syntactic and semantic hierarchies in Figures 1.31 and (81). One fact
about clause linkage that jumps out from them is that there are many more
semantic relations (31) than juncture–nexus types (14), and because of this
disparity, the semantic side of the IRH in Figure 1.32 contains the cluster
labels in (81), rather than the individual interclausal relations; where there
are ordering constraints within a cluster, they are specified. Moreover, few if
any languages have all of the syntactic relations, but presumably all lan-
guages can express the semantic distinctions. From this it follows that there
can be no one-to-one matching between the hierarchies; rather, there is a
many-to-one relationship between semantics and syntax. This is hardly
surprising; the analogy to grammatical relations is relevant here, as there
are many more thematic relations than grammatical relations, and so the
mapping from thematic to grammatical relations is likewise many-to-one.
The complexities of the many-to-one mapping between the syntactic and
semantics sides of the IRH are discussed in Ohori (Chapter 13) and Guerrero
(Chapter 14), as well as in Van Valin (2005, §6.6).33
The basic iconic principle entails that the tightest syntactic linkage type in
a language will instantiate the closest semantic relationship. Accordingly, if

Figure 1.32 The Interclausal Relations Hierarchy
causative [1] is realized syntactically, and if the language has constructions

instantiating nuclear cosubordination, then causative [1] will be expressed
via nuclear cosubordination, as in (60b) from Mandarin. However, it may be
the case that causative [1] is expressed exclusively by means of derivational
morphology. This is not a counterexample to the IRH, because, as mentioned
earlier, morphological constructions are even more tightly fused than
nuclear junctures, and this is in line with the principle of iconicity that
motivates the IRH. Another complexity alluded to earlier is the fact that
many verbs can license more than one juncture–nexus type, with each
coding a different interclausal semantic relation. Want, for example, can
be a psych-action predicate realized through core coordination, as in Bill
wants to visit Nancy tomorrow, or a propositional attitude predicate expressed
via core subordination, as in Bill wants very much for Sam to visit Nancy
tomorrow. (See Chapters 13, 14, Van Valin and Wilkins 1993, Van Valin
2005: 210–211.)
The IRH will be revisited in subsequent sections on semantic representa-
tion and on linking syntax and semantics in complex sentences.
1.3.4.4 English Causative Verbs

The cluster in (81b) contains the English verbs make, have, let and help, which
as a group have some intriguing properties and which pose certain chal-
lenges for the IRH. These properties are as follows. First, in the active voice
the infinitive following them cannot (make, have, let) or may not (help) be

marked by to. Second, make and help require to in the passive voice, while
have and let do not occur in the passive voice with or without it.34 This is
illustrated in (82).
(82) a. Sally made/had/let Tom (*to) clean the kitchen.

a′. Sally helped Tom (to) clean the kitchen.
b. *Tom was made/had/let/helped clean the kitchen by Sally.
b′. Tom was made/*had/*let/?helped to clean the kitchen by Sally.
It has long been observed that for many English speakers the two versions
in (82a′) can have different interpretations (e.g. Quirk et al. 1972; Bolinger
1975; Dixon 1984, 1991): in the to-less version Sally participated in the
cleaning with Tom, whereas in the version with to she need not have
done any of the cleaning but rather could have just given Tom some
cleaning supplies. In other words, with the bare infinitive Sally’s helping
and Tom’s cleaning temporally overlap, implying that she did some of
the cleaning, while in the to-infinitive version they need not overlap. This
illustrates a claim made in Van Valin and LaPolla (1997, §8.4.2) that units
expressing temporally overlapping states of affairs are often not joined by
any kind of linkage marker. In (82a) the unspecified causing/letting action
is simultaneous with the caused/permitted action, that is, they overlap
temporally, and this is consistent with the bare infinitive complement
predicate. This is the marked case; the default is for infinitives in English
to be marked by to.
The juncture–nexus type of these examples is not readily apparent. Two
factors suggest that this construction might be a nuclear juncture. First, it
implicates unmediated causation with make, and in many languages con-
structions with an explicit causative verb are nuclear junctures, for example
French faire ‘make, do’ þ infinitive, Italian fare ‘make, do’ þ infinitive,
Jakaltek a’a’ ‘give’ þ infinitive. Second, nuclear junctures almost never have
any kind of linkage marker. In English, for example, there is a clear differ-
ence between Nancy seems happy and Nancy seems to be happy, such that the
former is a nuclear juncture, a complex predicate formed in the syntax, and
the latter a core juncture, which requires to (and be) and is not a
complex predicate.
Evidence against (82a) being a nuclear juncture comes from reflexiviza-
tion.35 English reflexivization has often been referred to as ‘clause bound’,
that is, the controller and the reflexive have to be in the same clause, but
that is not correct in terms of the LSC. In a simple sentence with a single
core, this is an accurate description. However, if the clause contains two
cores, the controller must be a semantic co-argument of the reflexive, as
(83) a. Sami seems to have injured himselfi/him*i/j. Semantic co-arguments

b. Sami managed not to injure himselfi/him*i/j. Semantic co-arguments
c. Sami told Donnaj not to injure herselfj/her*j/k. Semantic co-arguments
d. Sami told Donnaj to help himself*i/himi/k. Not semantic co-arguments

In (83a) Sam and himself are both semantic arguments of the verb injure, and
despite the occurrence of Sam in a different core, the reflexive is obligatory.
The same is true in (83b): Sam is the shared argument between manage and
injure, and therefore Sam is a co-semantic argument of injure with the
reflexive anaphor. In (83c) Donna is in the core headed by tell but is the
shared argument with injure, and therefore Donna and herself are both
semantic arguments of injure. In the final example, however, Sam and himself
are not semantic arguments of the same verb and are in different cores, and
consequently Sam cannot be interpreted as the binder of himself and the
plain pronoun him must be used to refer to Sam.
The crucial sentence is given in (84a), with possible structures as both
nuclear (84b) and core (84c) junctures.36
(84) a. Bill made/had/let/helped Mary measure him(self ). Reflexive possible?

b. [CLAUSE [CORE Bill [NUC made ] Mary [NUC measure ] him(self ) CORE] CLAUSE]
c. [CLAUSE [CORE Bill [NUC made ] Mary CORE] [CORE [NUC measure ] him(self ) CORE]
CLAUSE]
If the constructions in (82) are nuclear junctures, then the structure would
be as in (84b), and because Bill and himself are within the same core,
reflexive binding between Bill and himself should be possible. On the other
hand, if they are core junctures, then the structure would be as in (84c),
and because Bill and himself are in different cores and are the semantic
arguments of different verbs, reflexive binding should not be possible. The
sentence in (84) is quite ungrammatical with the reflexive but compatible
with a plain pronoun, which shows that it is a core juncture, not a
nuclear juncture.
The next question concerns the nexus type. Daughter subordination can
be ruled out for the reasons given in Section 1.3.2, for example *It was clean
the kitchen that Sally made/had/let/helped Tom, and it is clearly not an adjunct in
the ad-core periphery. It must, therefore be coordination or cosubordina-
tion. For the reasons discussed with reference to (74) in Section 1.3.2, the
deontic modal scope test does not apply to this construction, but the scope
of negation test in (75) is applicable. It is illustrated in (85).
(85) a. The teacher didn’t make the students buy anything/??something.
a′. The teacher made the students buy *anything/something.
b. No teacher had/let the students buy anything/??something.
b′. The teacher had/let the students buy anything/something.
c. The teacher didn’t help the students (to) buy anything/??something.
c′. The teacher helped the students (to) buy *anything/something.
The results support a cosubordinate analysis, because in (85a–c) the core-

level negation in the first core licenses the NPI anything in the second core,
which is strongly preferred over the non-NPI something. In the non-negative
contexts in the primed examples, the NPI is predicted to be impossible,
which is the case in (a′) and (c′) but curiously not in (b′); this does not affect
the conclusion that a core-level operator is shared across both cores.

Additional support comes from the MEP and the LSC (Bohnemeyer and
Van Valin 2017). Bohnemeyer and Van Valin argue that the RRG notion of
the (verbal) core is the natural unit for expressing macro-events. Crucial for
this discussion is the fact that macro-events can have at most one temporal
positional modifier (TPM) (excluding relative clauses, which are RP-internal
and may be a separate macro-event), as illustrated in (86).
(86) a. Mary ate pancakes for breakfast [TPM today].

a′. [TPM *Yesterday], Mary ate pancakes for breakfast [TPM today].
b. [TPM This morning], Mary regretted kissing Bill at the party [TPM last night].
c. Mary asked her mother [TPM yesterday] to make pancakes for breakfast [TPM
today].
The simple sentence in (86a) has the MEP, because it allows only one TPM,
not two as in (a′). In the other two examples, a case of core-level daughter
subordination in (b) and core coordination in (c), neither construction has
the MEP, because each of them has two TMPs. The MEP, they argue, is
preserved in complex sentences only in core cosubordination. In (74) and
(75) try þ infinitive constructions were shown to be cases of core
cosubordination, and in (87), it can be seen that they have the MEP.
(87) *Sam tried [TPM in the morning] to fix the bicycle [TPM in the afternoon].
The following examples show that the causative [2] constructions have
the MEP.
(88) a. *Sam made Bill [TPM in the morning] fix the bicycle [TPM in the afternoon].
b. *Sam had Bill [TPM in the morning] fix the bicycle [TPM in the afternoon].
c. *Sam let Bill [TPM in the morning] fix the bicycle [TPM in the afternoon].
d. *Sam helped Bill [TPM in the morning] fix the bicycle [TPM in the afternoon].
d′. ?Sam helped Bill [TPM in the morning] to fix the bicycle [TPM in the afternoon].
The first four examples do not allow more than one TPM, supporting the
conclusion that the nexus type is cosubordination. The last example, while
odd, is not as bad as the first four, and this is consonant with the claim that
the help þ bare infinite form involves temporally overlapping actions while
the help þ to-infinitive does not necessarily involve temporally overlapping
actions, which renders the two TPMs less unacceptable. There is a clear
contrast in acceptability between the examples in (86b, c) and those in
(88), and taken together with the contrast between (75) and (85), they
strongly support the analysis of these constructions as being instances of
core cosubordination.
It should be noted that they are very unusual examples of core cosubordi-
nation, for two reasons. First, they are the only cases of core cosubordination
in which the ‘subject’ of the licensing verb is not the shared argument with
the linked verb, and second, they are the only ones which take bare infini-
tives rather than to-infinitives in the linked core, which is a feature of a
tighter morphosyntactic bond between the cores. This suggests that this is

the tightest version of core cosubordination, one step up from nuclear

junctures, as befits their causative semantics.
1.3.5 Complex RPs

It was stated at the beginning of Section 1.3 that the notion of ‘complex
sentences’ includes two components, clause linkage and complex RPs, and
the analysis of complex RPs borrows substantially from the theory of clause
linkage, which has received considerably more attention in RRG than the
theory of complex RPs. In particular, some of the juncture and nexus
concepts can be applied to RPs. It was shown in Section 1.2.5.3 that RPs have
a layered structure with three layers, each with a periphery: the nucleusR,
the locus of restrictive modification, the coreR, the domain of quantifica-
tion, and the RP-level, the locus of non-restrictive modification. The most
common complex RPs contain relative clauses, restrictive ones being housed
in the nucleusR periphery and non-restrictive ones in the RP-level periphery.
The RRG analysis of relative clauses is presented in París (Chapter 15), and
the discussion here will focus on the application of juncture and nexus to
complex RPs. It should be mentioned that considerably less work has been
done on this issue than on clause linkage and relativization.
Table 1.4 summarizes juncture and nexus in complex RPs.
All three nexus types are found at each level of juncture. Coordination at
the RP-level involves linked RPs with no operator dependence, which in
this case would be definiteness and deixis, and in this case RP-coordination
is realized by a conjunction construction, as are some of the other
juncture–nexus combinations. In the big dog and a little cat, each of the
linked units is a complete RP with its own RP-level operator. In the
cosubordinate linkage, however, there is a single definiteness operator
which has scope over both of the linked units. The second unit, little cat,
Table 1.4 Juncture and nexus in complex RPs
Level of juncture Nexus type Example
RP Coordination the big dog and a little cat

Cosubordination the big dog and little cat
Subordination (peripheral) the Pope, who is from Argentina, . . .
CoreR Coordination the big dogs and little cat
Cosubordination ang mga aso at pusa Tagalog
NOM PL dog and cat
‘the dogs and cats’
Subordination (daughter) the claim that the election was fraudulent
Subordination (peripheral) the many big dogs and little cats
NuclearR Coordination the two mugs of beer and glasses of wine
Cosubordination-1 the two barrels of beer and ale
Cosubordination-2 bottle opener, fire engine, duck hunter
Subordination (peripheral) the girl who sang, the big dogs and cats

cannot easily stand on its own, for example ??I brushed little cat, ??Little cat
meowed, unless it is interpreted as a name. Subordination at the RP-level
involves non-restrictive modification, and non-restrictive relative clauses
are housed in the RP-level periphery.
At the coreR-level the relevant operators are negation and number. In
coordination, each unit in the construction has independent operators at
the level of juncture; higher level operators are by definition shared. In the
big dogs and little cat, the first unit is plural and the second singular, showing
operator independence at the level of juncture. This also works for negation,
as in No men and three women survived the crash. This is trickier in
cosubordination, because the number morpheme is a bound affix and is
required. So in English it is not possible to mark number on one unit and
have it apply to another unit, that is, *the dogs and cat-__, where cat is
interpreted as plural. This is, however, possible in Tagalog (Kolmer 1998),
because the plural marker is not an affix but a free, albeit clitic-like,
element, as in the equivalent to the example just given, ang mga aso at pusa,
in which it has scope over both of the linked units; the RP cannot be
interpreted to mean ‘dogs and cat’, only ‘dogs and cats’. There is operator
dependence across the linked units, hence it is cosubordination. With
respect to negation, there appear to be forms in which negation is obligator-
ily shared across the units (e.g. neither dogs nor cats, *either dogs nor cats,
??neither dogs or cats). Both types of subordination are found at this level.
Nominals derived primarily from verbs of intentionality and speech in (81)
can take propositional complements, e.g. the belief/claim/rumour/assertion/fear
that . . ., which would be coreR daughter subordination. Quantificational
lexical modifiers, namely, quantifiers and numbers, occur in the coreR-level
periphery, as discussed in Section 1.2.5.3.
The nuclearR-level operator is nominal aspect, which includes nominal
and numerical classifiers and the mass/count distinction. In the table it is
represented by English measure words, which are a kind of numerical
classifier. NuclearR coordination is straightforward, and nuclearR
cosubordination-1 is analogous to it in that it involves sharing nominal
aspect (numerical classifiers) across the two nuclei instead of each nucleus
having its own classifier. NuclearR cosubordination-2, on the hand, is analo-
gous to the nuclear juncture in (60b) from Mandarin, in which the two
verbal nuclei form a complex predicate which takes a single aspect operator.
It consists of compounding two nominal nuclei to form compounds, for
example wood chopper, air conditioner, fire station, knife sharpener, computer
repairman, cat litter, dog poop, which have a single value in terms of the
mass/count distinction, which is a nuclearR-level operator, despite not
having an overt morphological manifestation. It is, rather, what Whorf
called a ‘covert category’, one that is signalled by its morphosyntactic
‘reactance’. In these compounds, the value is determined by the final word
in the group, which means there is operator dependence across the two
nuclei: wood is a mass noun, chopper is a count noun, and wood chopper is a

count noun; knife is a count noun, sharpener is a count noun, and knife
sharpener is a count noun; cat is a count noun, litter is a mass noun, and cat
litter is a mass noun. The other nuclearR-level juncture–nexus type is periph-
eral ad-subordination, which, as discussed earlier, is the locus of restrictive
modification.
There is a further class of complex RPs which clearly involve juncture–
nexus distinctions but have proven difficult to characterize: they are nom-
inalization of jussive and other verbs which take an infinitive complement,
for example the (commander’s) order to the troops to attack the fortress, the
promise to the workers from the boss to increase their wages/the boss’s promise to
the workers to increase their wages, the request from the teacher to the students to
work quietly/the teacher’s request to the students to work quietly. There is a shared
argument in these constructions, just as in the corresponding verbal
predications discussed in Section 1.3.2, and accordingly these would be
non-subordinate core junctures. By virtue of the deontic modal scope test,
it was determined that the verbal predictions are core coordination, but
there are no comparable tests applicable to these complex RPs to distin-
guish the nexus types. It seems reasonable, then, as a preliminary analysis,
to categorize them as coreR coordination, based on the strong parallels
with the corresponding verbal predications. More research is needed on
these constructions.
1.4 Semantic Representation
1.4.1 Introduction
In Figure 1.1 there are two representations, one syntactic and the other
semantic, and the nature of the syntactic representation was explicated in
Sections 1.2 and 1.3. The spotlight in this section is the semantic representa-
tion, which contains two major parts: the system of lexical representation,
on the one hand, and the theory of semantic roles, on the other. The basics
of each will be presented, as more extensive discussion can be found in
Mairal Usón and Faber (Chapter 3) and Kailuweit (Chapter 4). Then some
extensions of these ideas will be presented.
1.4.2 Lexical Representation

RRG has long had a very rich lexical representation, based on lexical decom-
position, in contrast to other approaches which represent the semantics of
verbs by a list of thematic relations.
1.4.2.1 Lexical Decomposition

Foley and Van Valin (1984) adopted the system of lexical decomposition put
forward in Dowty (1979), using Dowty’s formalization of the Aktionsart
categories originally proposed in Vendler (1967). There were four categories,

namely states, activities, achievements and accomplishments, and they

were formalized as in (89), and the representations are known as logical
structures (LSs).
(89) a. State: predicate′ (x) or (x, y), e.g. dead′ (rat) ‘The rat is dead’, know′ (Bill,
French) ‘Bill knows French’
b. Achievement: BECOME predicate′ (x) or (x, y), e.g. BECOME dead′ (rat) ‘The
rat died’, BECOME know′ (Bill, French) ‘Bill learned French’
c. Activity: DO predicate′ (x) or (x, y), e.g. DO dance′ (Sally), ‘Sally danced’, DO
eat′ (Bill, pizza) ‘Bill ate pizza’.
d. Accomplishment: ϕ CAUSE ψ (where ϕ is normally an activity predicate
and ψ an achievement predicate, e.g. [DO predicate′ (x)] CAUSE [BECOME
predicate′ (y) or (y, z)], e.g. [DO do′ (cat)] CAUSE [BECOME dead′ (rat)] ‘The
cat killed the rat’,
[DO do′ (Sally)] CAUSE [BECOME know′ (Bill, French)] ‘Sally taught French
to Bill’37
There are several appealing features of this approach. First, the main lexical
semantic content of verbs is represented by state and activity predicates
only, with change-of-state (achievement) and caused-change-of-state (accom-
plishment) verbs being derived via the addition of elements like BECOME
and CAUSE. Second, the state $ achievement $ accomplishment pattern is
directly reflected in the derivational morphology of verbs in many lan-
guages. Third, there is a set of syntactic and semantic tests for identifying
the class to which a particular use of a verb can be assigned (see Van Valin
2005, §2.1.1). One of the serious complications in the analysis of verb seman-
tics is that a given predicate, for example cool, can be used in more than one
way, as in The soup is cool (state), The soup cooled (achievement) and The ice cooled
the soup (accomplishment), and consequently what one is identifying in most
cases is the interpretation of a predicate in a particular sentence. The
question of how to analyse such a predicate is left open: one could claim
that there are three entries in the lexicon for the verb form cool (cool1 ¼ state,
cool2 ¼ achievement, cool3 ¼ accomplishment), or one could claim that there
is only one entry, cool ¼ cool′ (x), with the other two forms derived by means
of lexical derivational rules. RRG has always favoured the latter approach
(e.g. Van Valin 2012a).
The decomposition system was basically the same from 1984 through
1997, when Van Valin and LaPolla introduced a major revision of it. The
primary features of the revision involved returning to the original distinc-
tions as proposed in Vendler (1967)38 and factoring out causation as an
independent parameter: all of the Aktionsart types have causative and non-
causative versions. With respect to state predicates, more distinctions
among subtypes were recognized, following Schwartz (1993). Activity
predicates were marked with do′, in order to distinguish them from states
(e.g. see′ (Mary, child) ‘Mary saw the child’ [state] vs. do′ (Mary, [see′ (Mary,
child)]) ‘Mary watched the child’ [activity]). Change-of-state predicates went
from one category, achievements, to three: achievements (punctual

changes of state, marked by INGRessive (e.g. INGR popped′ (balloon) ‘The

balloon popped’)), accomplishments39 (non-punctual changes of state with
an inherent endpoint, marked by BECOME (e.g. BECOME melted′ (ice) ‘The
ice melted’)), and processes (non-punctual changes of state or location
without an inherent endpoint, marked by PROC (e.g. PROC melt′ (ice)
‘The ice was melting’)).40 In fact, BECOME can be decomposed into PROC
& INGR, ‘&’ meaning ‘and sequential’, so The ice melted would be PROC melt′
(ice) & INGR melted′ (ice), which can be abbreviated as ‘BECOME melted′
(ice)’ (Van Valin 2005: 44). Both activities and processes have temporal
duration, and both lack inherent endpoints (they are atelic); they differ
in that activities are dynamic and processes are not, for example The house
shook violently/intensely during the earthquake vs. *The ice melted violently/
intensely during the storm.
Processes with endpoints are (process) accomplishments; what are activ-
ities with endpoints? Such predications exist: Bill jogged in the park (no end-
point) vs. Bill jogged to the park (endpoint), Max ate pizza (no endpoint) vs. Max
ate (up) the pizza (endpoint). Activities with endpoints are termed active
accomplishments in RRG, and unlike in other approaches, they are
treated as a distinct category, as argued for in detail in Van Valin (2018a).
It has been objected that they do not constitute a distinct category, because
there are supposedly no lexical active accomplishments, but this is incor-
rect: devour is a lexical active accomplishment, as is enter (Bill jogged into the
room ¼ Bill entered the room (jogging)). Active accomplishments always involve
an incremental theme or incremental path. Incremental themes are associ-
ated with consumption and creation verbs, while incremental paths are a
feature of motion verbs. The three types of verbs have slightly different LSs
(from Van Valin 2018a).
(90) a. Consumption predicates, e.g. devour

[do′ (x, [eat′ (x, y)]) ^ PROC consume′ (y)] & INGR consumed′ (y)
b. Creation predicates, e.g. write (a poem)
[do′ (x, [write′ (x, y)]) ^ PROC create′ (y)] & INGR exist′ (y)
c. Motion predicates, e.g. enter
[do′ (x, [move′ (x)]) ^ PROC cover.path.distance′ (x, (y))] & INGR be-LOC ′ (z, x)
The LS in (90a) has the interpretation ‘x eats y, which simultaneously

undergoes a process of consumption, both of which terminate and lead to
the result that y is consumed.’ The one in (90b) should be read as ‘x writes y,
which simultaneously undergoes a process of creation, which terminates
and leads to the result that y exists’. The final one signifies ‘x moves and
simultaneously effects a process of covering distance, which terminates,
which leads to the result that x comes to be located at/in/on z’.
Osswald (2021) presents a critical review of the decomposition system in
RRG, and two of his points are particularly relevant to the LSs proposed for
both process and active accomplishments. The first concerns the nature of
the predicate in the process LS, PROC predicate′. The decomposition of the

process accomplishment melt given above is PROC melt′ & INGR melted′,
and the question arises, if melted′ is a state, what is melt′? The system has
only two types of basic predicates, states and activities, and melt′ fits into
neither category. The same issue appears in the active accomplishment LSs
in (90a, b) with PROC consume′/create′ (y). Osswald notes that the intended
meaning in (90) is being.consumed′ and being.created′.
Processes are non-punctual changes of state or location, and therefore the
argument-bearing lexical predicate at the heart of the representation must
be either one of state or condition, or one of location. Many processes do not
have an incremental theme argument which is being consumed or created
or an incremental path; rather, they involve the position of the referent on
a scale. This is particularly true of what Dowty (1979) termed ‘degree
achievements’ such as cool, warm, widen, narrow, grow, melt, freeze, redden
(all [M]-intransitive), all of which assume a scale of some kind. For example,
there is a temperature scale with ‘hot’ at the top and ‘cold’ at the bottom, as
can be seen in the contrast between cool down/*up vs. warm up/*down. Osswald
notes that as a process the soup cooled means ‘the soup became cooler’, not
‘the soup became cool’. Hence the incremental theme of these verbs is
associated with movement on a scale rather than consumption or creation.
This can be represented as PROC becoming.higher/lower.on.[α]scale′ (x).
These representations carry over to their [M]-transitive causative
counterparts.
Nothing has been said about motion active accomplishments, which
involve an incremental path rather than an incremental theme. Crucially,
an incremental theme undergoes a change of state, but an incremental
path does not; it measures out the distance the x-argument covers, which
means the proposed predicate, covering.path.distance′, cannot be a predi-
cate of state or condition. It is, rather, a change-of-location predicate,
which can co-occur with motion activity predicates, as well as pure process
predicates like fall, as in He fell fifty metres into the water. In motion active
accomplishments there is an activity simultaneous with a process, the
former characterizing the motion itself and the latter expressing the
distance along the incremental path, whereas in the example with fall
there is an uncontrolled change of location in a specific direction covering
a specified distance before reaching a final location. Hence the LS for an
active accomplishment like stumble into the room or enter the room stumbling
should not be (90c) but rather [do′ (x, [move′ (x)]) ^ PROC covering.path.
distance′ (x, (y))] & INGR be-LOC ′ (z, x).41 Consequently, the LS for motion
active accomplishments parallels the LSs for creation and consumption
predications.42 On the other hand, the LS for the example with fall would
be [PROC moving.downward′ (x) ^ PROC covering.path.distance′ (x, (y))] &
INGR be-LOC ′ (z, x), where ‘moving′’ means ‘changing location’. Moving′
differs from do′ (x, [move′ (x)]), in that it necessarily does not involve any
effort on the part of the referent of its argument, is primarily associated
with direction rather than manner and is not subject to the agentivity

implicature, even with a human referent (see §1.4.3.1), whereas do′ (x, [move
′ (x)]) differs on all three of these characteristics.
The second issue raised by Osswald is what he calls the ‘and-then anomaly’.
He points out that representations like ‘PROC melt′ (x) & INGR melted′ (x)’
don’t accurately represent the situation. This one means ‘there was a process
of melting, and then after the process ended there was a transition to the
state of being melted’, but the state of being melted does not come about
after the process ended; rather, it is the final stage of the process itself.
Similarly, in active accomplishments like Sam devoured the pizza, it is not the
case that the activity of eating and the process of being consumed termin-
ated and then the pizza was consumed; the pizza was consumed at the
termination of the eating, not after it. A possible solution to this anomaly is
to introduce a function FIN ‘final stage of a process or action’ and change the
connector from & ‘and then’ to ^ ‘and simultaneously’, which is adapted
from Osswald’s proposal. Thus The ice cream melted would be PROC becoming.
lower.on.[solid$liquid]scale′ (ice cream) ^ FIN melted′ (ice cream), mean-
ing ‘the ice cream underwent a process of becoming lower on the “solid-to-
liquid” scale and the final stage is that the ice cream is melted’. For Sam
devoured the pizza, the LS would be do′ (Sam, [eat′ (Sam, pizza)]) ^ PROC being.
consumed′ (pizza)] ^ FIN consumed′ (pizza), meaning ‘Sam eats pizza and at
the same time the pizza undergoes a process of consumption whose final
stage is the pizza is consumed’.
The introduction of the FIN function offers a way to capture the contrast
between quantized and non-quantized changes of state. Following Beavers
(2013), Bentley (2019) argues that the previous decomposition system does
not express the difference between quantized change-of-state predicates
like die, which lexicalize a specific result state, and non-quantized change-
of-state predicates, which are of two types: those like melt and fill, which
entail only that a specific result state (e.g. being liquid, being full) might
but need not necessarily be reached, and those like widen or shorten which
have no specific final state. The following represents the result state
of the two subclasses of non-quantized change (D. Bentley, personal
communication):
(i) melt: PROC becoming.lower.on.[solid $ liquid]scale′ (ice cream) ^ FIN

melted.by.some.amount′ (ice cream) (the ice cream was not melted to
begin with and now is melted by some amount, which may be at the end
of the scale, i.e. the ice cream was completely melted)
(ii) widen: PROC becoming.higher.on.[wide $ narrow]scale′ (crack) ^ FIN
wide.by.larger.amount′ (crack) (the crack was wide by some amount and
now it is wide by a larger amount)
This notation yields the necessary contrast with verbs of quantized change,
which entail the reaching of a specific final state (e.g. FIN dead′ (x)).
The final addition to the array of Aktionsart types is semelfactives
(Smith 1997), which are punctual events with no change of state or result

state; they contrast primarily with achievements, which are punctual

changes of state with a result state. For example, it is possible to say Sam
flashed the light, and then he re-flashed it again but not *Sam popped the firecracker,
and then he re-popped it again. Flashing a light does not cause it to undergo a
change of state, and consequently it can be flashed again. In contrast,
popping a firecracker causes it to undergo a change of state, and accordingly
it cannot be popped again. Hence flash in this example is a (causative)
semelfactive, while pop is a (causative) achievement. The LS for semelfactives
is SEML do′ (x, [pred′ (x, (y))]). Because they are events, they are related to
activities rather than states.
The lexical representations for the Aktionsart categories are given in
Table 1.5.43
Examples of each type are given in (91).
(91) a. STATES
The window is shattered. shattered′ (window)
John saw the picture. see′ (John, picture)
b. ACTIVITIES
The children cried. do′ (children, [cry′ (children)])
Carl ate pizza. do′ (Carl, [eat′ (Carl, pizza)])
c. ACHIEVEMENTS
The window shattered. INGR shattered′ (window)
The climber reached the summit. INGR be-at′ (summit, climber)
d. SEMELFACTIVES
Mary coughed. SEML do′ (Mary, [cough′ (Mary)])
Dana glimpsed the picture. SEML do′ (Dana, [see′ (Dana, picture)])
e. ACCOMPLISHMENTS
The ship sank. BECOME sunken′ (ship)
Mary learned French. BECOME know′ (Mary, French)
f. ACTIVE ACCOMPLISHMENTS
Carl ate the pizza. do′ (Carl, [eat′ (Carl, pizza)]) ^ PROC being.consumed′
(pizza) ^ FIN consumed′ (pizza)
Chris ran two miles to the park. do′ (Chris, [run′ (Chris)]) ^ PROC covering.path.distance′
(Chris, two miles) ^ FIN be-at′ (park, Chris)
g. CAUSATIVES
The dog scares the boy. [do′ (dog, Ø)] CAUSE [feel′ (boy, [afraid′])]
The submarine sank the ship. [do′ (submarine, Ø)] CAUSE [BECOME sunken′ (ship)]
The cat popped the balloon [do′ (cat, Ø)] CAUSE [INGR popped′ (balloon)]
The conductor flashed the light. [do′ (conductor, Ø)] CAUSE [SEML do′ (light, ([flash′ (light)])]
Felix rolled the ball. [do′ (Felix, Ø)] CAUSE [do′ (ball, [roll′ (ball)])]
Mary fed the pizza to the child. [do′ (Mary, Ø)] CAUSE [do′ (child, [eat′ (child, pizza)]) ^
PROC being.consumed′ (pizza) ^ FIN consumed′ (pizza)]
The causative LSs typically have an activity predicate as the first argument
of CAUSE,44 and it is quite common cross-linguistically for that activity to be
unspecified, as in all of the examples in (91). This unspecified activity
has heretofore been represented as ‘[do′ (x, Ø)]’. An alternative way of
representing this activity but with more semantic content is the

Table 1.5 Lexical representations for Aktionsart classes
Verb class Logical structure
STATE predicate′ (x) or (x, y)

ACTIVITY do′ (x, [predicate′ (x) or (x, y)])
ACHIEVEMENT INGR predicate′ (x) or (x, y), or
SEMELFACTIVE SEML do′ (x, [predicate′ (x) or (x, y)])
PROCESS PROC being.consumed′/created′ (y) or (x, y)
PROC becoming.higher/lower.on.[α]scale′ (x)
PROC moving(.direction)′ (x)
PROC covering.path.distance′ (x, (y))
ACCOMPLISHMENT BECOME predicate′ (x) or (x, y), where
BECOME ¼ PROC ^ FIN
ACTIVE ACCOMPLISHMENT
do′ (x, [pred′ (x, y)]) ^ PROC being.consumed′ (y)] ^ FIN consumed′ (y)
do′ (x, [pred′ (x, y)]) ^ PROC being.created′ (y)] ^ FIN exist′ (y)
do′ (x, [pred′ (x)]) ^ PROC covering.path.distance′ (x, (y)) ^ FIN be-loc′ (z, x)
CAUSATIVE α CAUSE β, where α, β are LSs of any type
following: [do′ (x, [affect′ (x, y))] CAUSE [BECOME/INGR pred′ (y)]], which
signifies ‘x does something unspecified which affects y causing y to undergo
a change of state’. This has the advantage that it ties the unspecified causing
activity to the affected participant in the second LS in the causative LS. For
example, the LS for The burglar smashed the window would be ‘[do′ (burglar,
[affect′ (burglar, window))] CAUSE [INGR smashed′ (window)]’. Another alter-
native along the same lines is ‘[do′ (burglar, [act.on′ (burglar, window))]
CAUSE [INGR smashed′ (window)]]’. For the examples in (91g) any of the
three would be appropriate, but there are some interesting cases where
either of the latter two would be better. They involve predicative
prepositional phrases.
In Section 1.2.5.1 the distinction between predicative and non-predicative
PPs was introduced (see also Ibáñez Cerda, Chapter 10). Non-predicative
adpositions are oblique core arguments of the predicator in the nucleus of
the core, and they are assigned by rule or constructionally and are not directly
represented in the LS of the predicator. Predicative adpositions, on the other
hand, are predicates and are represented as such in the semantic representa-
tion of the sentence. In a sentence like Pam saw Felix in the library after the
lecture, the LS would be be-after′ (lecture, [be-in′ (library, [see′ (Pam, Felix)])]), in
which after the lecture (be-after′ (x, y)) and in the library (be-in′ (x, y)) are realized
as adjunct PPs in the core-level periphery. The first argument is the spatial or
temporal location of the event represented by the second argument. The
default situation is that non-predicative PPs occur in the core as oblique core
arguments and predicative PPs occur in a periphery as an adjunct.
There are, however, two situations in which this default distribution of
PPs does not hold. The first is the by-PP in a passive construction; it is a non-
predicative PP because it is constructionally motivated and is not repre-
sented in the LS, but it occurs in the core-level periphery. It is a case of a

non-predicative PP in a periphery rather than the core. The second situation

involves verbs like put, which have a requirement for a locative element but
do not specify what it should be. In the case of English put, there is a range of
possibilities, all contributing to the meaning of the LS: a locative demon-
strative (put it here/there), an intransitive preposition (put it down/away), or
a full predicative PP (put it behind the chair/on the table/under the bed,
etc.). These locative expressions contribute to the meaning of the clause and
are represented as predicates in the LS. Thus, they satisfy a valence require-
ment of a predicate and occur in the core, like an argument, but they are
also independent semantic units which are not determined by the predicate
and contribute semantically, like an adjunct. Hence they are termed
argument-adjuncts.
In a sentence like Max put the book in the box, the RP the book is an argument
of both put and the PP in the box. The LS for put would be [do′ (x, Ø)] CAUSE
[INGR be-LOC ′ (. . ., z)], where ‘be-LOC ′’ can be realized by any of the three
possibilities given above, and the usual LS for this sentence would be [do′
(Max, Ø)] CAUSE [INGR be-in′ (box, book)]. This represents the book primarily
as an argument of be-in′, the predicative preposition, and the fact that Max
does something to the book to cause it to move into the box is not overtly
represented. Here is where the alternative formulations of the unspecified
causing activity have an advantage: they both explicitly represent the argu-
ment sharing between the verb and the preposition, that is, [do′ (Max, [act.
on′ (Max, book)])] CAUSE [INGR be-in′ (box, book)]. This argument sharing
between the two predicates distinguishes pure adjunct PPs, which take the
LS of the core as a whole as an argument, from argument-adjuncts, which
take one of the arguments in the LS of the core as an argument.
Adjectives and adverbs are also represented as predicates in the semantic
representation. Following Schwartz (1993), the LS for attributive predica-
tions is ‘be′ (x, [pred′])’ (e.g. be′ (grass, [green′]) for The grass is green). The
semantic representation for attributive modification is similar, for example
Bill saw the green grass would be see′ (Bill, [be′ (grass, [green′])]); the underlin-
ing of grass signals that as head of the argument RP it functions simultan-
eously as an argument of the embedded attributive predication and as an
argument of the main predicate see. This same double function is found in
possessive constructions, for example Bill saw Mary’s car would have the LS
see′ (Bill, [have′ (Mary, car)]), and Bill saw Mary’s new car would have two
embedded propositions in the second argument position, that is, see′ (Bill,
[have′ (Mary, [be′ (car, [new′])])]). In this LS car functions as an argument in
three predications.45
Three points need to be mentioned. First, the fact that the attributive LS
contains ‘be′’ in no way implies that a language necessarily has a copular
verb analogous to English be, nor is it a claim that there is some kind of
underlying copular verb. It is a part of the attributive LS, nothing more. It is
analogous to CAUSE: it represents causation in a causative LS, but it does not
imply that every language has a causative verb like French faire or a

causative morpheme like Lakhota ka- ‘cause by striking’. Second, the predi-
cate in an attributive LS can also occur in another type of LS. The adjective
black can occur as an attributive predicate, as in be′ (door, [black′]) The door is
black, an attributive modifier, as in like′ (Mary, [be′ (door, [black′])]) Mary liked
the black door, and also in a causative change of state LS, as in [do′ (fire, [act.
on′ (fire, door)])] CAUSE [BECOME black′ (door)] The fire blackened the door. In
this sentence black′ (door) is the result state of a causative accomplishment,
not an attribute, and accordingly the representation is different. Third,
attributive predicates can have second arguments, for instance Sam is proud
of his son, Nancy is angry at her husband, and the former would be represented
as be′ (Sam, [proud.of′ ([have.as.kin′ (3sgM, son)])]).
Adverbs have a simple LS, as they are one-place predicates, for example Bill
carefully closed the door slowly yesterday would be yesterday′ ([do′ (Bill, [careful′
(act.on′ (Bill, door))])] CAUSE [slow′ (BECOME closed′ (door))]). As this
example illustrates, adverbs may have the whole LS as their argument or a
subpart, depending on the level and scope of modification, as discussed in
Section 1.2.4.2.
The semantic representation of nominals does not treat them as predi-
cates, and their semantic representation is based on the qualia analysis
proposed in Generative Lexicon Theory (Pustejovsky 1995). The qualia them-
selves are represented in terms of the decomposition system used in RRG,
which differs from that used by Pustejovsky. See Van Valin (2005, §2.3) for
further discussion.
Operators have a place in the semantic representation of both RPs and
clauses. The RP operators and the semantic representation for Sally’s two red
parrots in Figure 1.18 are in (92).
(92) a. ⟨DEF ⟨DEIX PROX/DISTAL ⟨NEG ⟨NUM SG/DL/PL ⟨NASP COUNT/MASS ⟨[N]⟩⟩⟩⟩⟩⟩
b. ⟨DEF þ ⟨NUM PL ⟨NASP COUNT ⟨[have′ (Sally, [be′ ([be′ (parrot, [red′])]) [two′])])]⟩⟩⟩⟩
The operators over the clause are summarized in (93a) and the semantic
representation for The dogs may be barking at Sally’s two red parrots is given in
(93b). These semantic representations must ultimately be interpreted within
a formal semantic framework.
(93) a. ⟨IF DECL ⟨EVID Ø ⟨TNS PAST ⟨STA IRR ⟨NEG ⟨MOD ABLE ⟨DIR Ø ⟨ASP IMPF ⟨[LS]⟩⟩⟩⟩⟩⟩⟩⟩⟩
⟨DEF þ ⟨NUM PL ⟨NASP COUNT ⟨[dog (x)]⟩⟩⟩⟩
\
b. ⟨IF DECL ⟨TNS PRES ⟨STA PSBL ⟨NEG – ⟨ASP PROG ⟨[do′ (x, [bark.at′ (x, y)])]⟩⟩⟩⟩⟩⟩
/
⟨DEF þ ⟨NUM PL ⟨NASP COUNT ⟨[have′ (Sally, [be′ ([be′ (parrot (y), [red′])]) [two′])])]⟩⟩⟩⟩
1.4.2.2 Representing the Semantics of Clause Linkage

The idea of using the system of lexical decomposition introduced in the
previous section to formally characterize the interclausal semantic

relations of the IRH in (81) was originally suggested in Ohori (2001), and
Van Valin (2005: 207–208) used it to formalize the version of the IRH
assumed then. Before the revised system is presented, a few notes of
clarification are in order. First, ‘VERB ′ ’ stands for the class of predicates
of this type; thus, know′ is the verb know but KNOW ′ symbolizes the class of
cognition verbs. Causative [1] has the same LS as the lexical causative verbs
in (91g) and need not involve a causative verb like English make, German
lassen or French laisser, whereas causative [2] typically does feature an expli-
cit causative verb; causative [3] always involves two verbs, one of which may
be explicitly causative, such as English cause, force, or not, for example
English teach in teach Sam to fix bicycles. ‘pred′ (x, [LS . . . x . . .])’ means that
the participant denoted by x is involved in both the matrix and embedded
LSs, as in (66). ‘pred′ (x, [LS . . . y . . .])’ also signals that the participant denoted
by y is involved in both the matrix and embedded LSs but in a different way,
as in (67).
(94) a. Single actions

1. Causative [1] [subevent1] CAUSE [subevent2]
2. Modifying sub-actions
i. Method do′ (x, [pred1′ (x, y)] CAUSE [INGR/BECOME state.pred2′ (y)])
ii. Manner do′ (x, [MOTION ′ (x)] . . . ^ [MANNER . OF . MOTION ′ (x)])
3. Phase do′ (x, [ONSET ′ (x, [LS . . . x . . .])], do′ (x, [CONTINUE ′
(x, [LS . . . x . . .] )], do′ (x, [TERMINATE ′ (x, [LS . . . x . . .])]
b. Multiple actions
4. Simultaneous
i. Non-motion do′ (x, [pred1′ (x, y)]) ^ do′ (x, [pred2′ (x, y)])
ii. Motion do′ (x, [MOTION ′ (x)] . . . ^ [pred2′ (x, (y))])
iii. Position do′ (x, [STANCE ′ (x)] ^ [pred2′ (x, (y))])46
5. Sequential do′ (x, [pred1′ (x, y)]) & do′ (x, [pred2′ (x, y)])
6. Causative [2]
i. Unmediated [do′ (x, [MAKE ′ (x, y)])] CAUSE [. . . pred2′ (y, (z))])
ii. Voluntary [do′ (x, [HAVE ′ (x, y)])] CAUSE [do′ (y, [pred2′ (y, (z))])])
iii. Facilitating [do′ (x, [LET ′ (x, y)])] CAUSE [ … pred2′ (y, (z))])
iv. Assisting [do′ (x, [HELP ′ (x, y)])] CAUSE [do′ (y, [pred2′ (y, (z))])])
c. Endeavour
7. Attempt do′ (x, [ATTEMPT ′ (x, [LS . . . x . . .])])
8. Succeed do′ (x, [SUCCEED ′ (x, [LS . . . x . . .])])
9. Failure do′ (x, [FAIL ′ (x, [LS . . . x . . .])])
d. Intentions
10. Refusal do′ (x, [REFUSE ′ (x, [LS . . . x . . .])])
11. Psych-action MENTAL . DISPOSITION ′ (x, [LS . . . x . . .])
12. Purposive want′ (x, LS2) ^ DO (x, [ [LS1] ◇CAUSE [LS2]])
e. Bringing about
13. Causative [3] [do′ (x, [VERB 1 ′ (x, y)])] CAUSE [. . . pred2′ (y, (z))])
14. Jussive [do′ (x, [SAY ′ (x, y)])] CAUSE [MENTAL.DISP′ (y, [LS . . . y . . .])]
15. Permissive [do′ (x, [PERMIT ′ (x, y)] CAUSE [ … pred2′ (y, (z))])
16. Injunctive [do′ (x, [STOP ′ (x, y)])] CAUSE [NOT . . . pred2′ (y, (z))])

f. Perception
17. Direct PERCEIVE′ (x, [LS . . . y . . .])
18. Indirect PERCEIVE′ (x, [LS])
g. Intentionality
19. Prop. attitude BELIEVE′ ((x,) [LS])
20. Cognition KNOW′ (x, [LS])
21. Emotion FEEL′ (x, [LS])
h. Speech
22. Indirect discourse do′ (x, [say′ (x, [LS ⟨TNS . . . ⟩])])
23. Direct discourse do′ (x, [say′ (x, [LS ⟨IF . . . ⟩])])
i. Locational
24. Space be-LOC′ ([LS1], [LS2])
25. Time be-TEMP′ ([LS1], [LS2])
j. Circumstances
26. Reason [LS1] BECAUSE′ [LS2]
27. Conditional [LS1] [LS2]
28. Concessive [LS1] IN.SPITE.OF′ [LS2]
k. Temporality
29. Simultaneous [LS1] ^ [LS2]
30. Sequential [LS1] & [LS2]
31. Unordered [LS1] þ [LS2]
These representations make it possible to give semantic representations for

complex sentences involving clause linkage.
1.4.2.3 Representing Constructional Meaning

The system of lexical decomposition is designed to represent the meaning of
lexical items and their combinations in sentences. There are instances in
which the meaning is not a function of the lexical items but is constructional
in nature. The English resultative construction, for instance Mary painted the
door black is an example of this; there is no lexical item coding causation, just
Mary painted the door and the secondary predicate black, and the combination
leads to the causative interpretation. The same is true in the Mandarin
construction in (60b); the lexical verbs are dǎ ‘beat’ and sı̌ ‘die’, and the
combination has the meaning ‘beat to death’. This raises the question, how
is constructional meaning to be represented? Should it be represented the
same as lexical meaning? In (94) causation is signalled the same way using
‘CAUSE’, regardless of whether it is lexical or constructional. Other cases of
constructional meaning involve more than adding a causative meaning to
two co-occurring predicators; a prime example is possessive predications in
languages which lack a verb like have. Examples are given in (95)–(97).
(95) a. U menja est’ knig-a. Russian

at 1sg.gen exist book-nom
‘I have a book.’ [Literally: ‘at me exist a book’]
b. U menja net knig-i.
at 1sg.gen not.exist book-gen
‘I don’t have a book.’ [Literally: ‘at me not.exist a book’]

(96) a. Bayan nhangu wunaa. Guugu Yimidhirr

house 3sg.dat exist
‘He/she has a house’ [Literally: ‘his/her house exists’, ‘a house exists to
him/her’]
b. Mayi gabiir-bi guya.
food girl-dat not.exist
‘The girl has no food.’ [Literally: ‘the girl’s food does not exist’, ‘food
doesn’t exist to the girl’]
(97) a. May libro ang babae. Tagalog

exist book nom woman
‘The woman has a book.’ [Literally: ‘a book exists the woman’]
b. Wala-ng libro ang babae.
not.exist-lnk book nom woman
‘The woman doesn’t have a book.’ [Literally: ‘a book doesn’t exist the woman’]
None of these languages has a lexical verb expressing possession akin to

English have, French avoir, Lakhota yuhá or Mandarin yǒu.47 In all of them an
existential verb is crucial to the expression of possession and a negative
existential verb for lack of possession. This is perhaps not surprising, given
the close connection between existential expressions and those of having; in
French, for example, the verbal part of the existential il y a is a ‘has’, and in
Mandarin yǒu (same character) is also used in existential constructions, for
example Zhuō shàng yǒu yı̄ běn shū [table on exist one CL book] ‘There is a book
on the table’ vs. Wǒ yǒu sān běn shū [1sg have three CL book] ‘I have three
books’.48
In (95) from Russian, the possessor is the genitive object of the locative
preposition u ‘at’, and the possessed argument is formally the subject in the
nominative case, although the predicate est’ ‘exist’ is impersonal and does not
agree with the nominative RP. When the verb is negated, as in net (< ne ‘not’ þ
est’ ‘exist’), the single argument appears in the genitive case. In Guugu
Yimidhirr, an Australian language spoken in northern Queensland (Haviland
1979), an alienable possessor in an RP is in the dative case, as in bayan nhangu
[house 3sgdat] ‘his/her house’, mayi gabiir-bi [food girl-dat] ‘the girl’s food’. In
a clause, the possessor can be analysed as RP-internal, yielding the first
readings given above, or as a core argument, yielding the second readings.
The latter parallels the Russian pattern, namely ‘a book exists at me’ vs. ‘a
house exists to him/her’. The Tagalog pattern, while similar, differs in two
striking respects: the possessor is neither dative nor oblique but nominative,
and the possessed is obligatorily not case-marked at all.
The issue concerns the LS for these predications. One possibility is to say
that since all of them are used to express the equivalent of constructions
with have, they should all get the LS have′ (x, y), just like in languages with a
possessive verb like have. While this captures the semantic commonality
among the forms, it will engender a lot of arbitrary linkings between syntax
and semantics in the languages, since other two-place state predicates
follow the standard linking pattern(s) for such predicates in the language.

Because of this problem, it might be suggested that the LSs should reflect the
properties of the specific constructions. On this view, the LS for (95a) would
be something like be-at′ (1sg, [exist′ (knig-)]), which can be readily accommo-
dated in terms of the regular Russian linking and case assignment rules;
(96a) would be exist′ ([have′ (3sg, bayan)]), where the possessor can be linked
as an RP-internal possessor or linked via ‘possessor raising’ as an argument
in the core of the clause; and (97a) would include exist′ (libro), but it’s not
clear how the possessor is related to it, as it seems to be simply juxtaposed to
it (i.e. ([exist′ (libro)] þ babae)), which is unsatisfactory. Also unsatisfactory is
the semantic interpretation of these LSs, since they should mean the same as
their counterparts in the languages mentioned above.
In Latrouite and Van Valin (2014) a solution to this problem is pro-
posed: constructional meaning is represented by the same basic decom-
positional system as lexical meaning, but the notation for each is
distinctive. Lexical meaning continues to be represented as before, with
elements in the semantic metalanguage in boldfaceþprime (e.g. have′).
Constructional meaning, on the other hand, will be represented by small
capsþprime (e.g. have′). In building representations of constructional
meaning, the components INGR, PROC, BECOME and SEML have the same
function as in the representation of lexical meaning, but CAUSE must now
be treated differently in the two representational notations, because it can
be lexically realized, as in the examples in (91g), still symbolized by CAUSE,
or it can be constructionally induced, as in the examples discussed earlier,
symbolized by cause in small caps, for example [do′ (1sg, [beat′ (1sg, shé)])]
cause [INGR dead′ (shé)] for (60b) from Mandarin, [do′ (Mary, [paint′ (Mary,
door)])] cause [INGR black′ (door)] for Mary painted the door black. Accord-
ingly, the LS in (94a1) should have CAUSE*, which indicates both are
possible, and (94a2i) should have cause rather than CAUSE, since it typic-
ally does not contain a lexical coding of causation, unlike the causative [2]
relations in (94b6).
With respect to the possessive predications in (95)–(97), the LSs would be
be-at′ (1sg, [exist′ (knig-)]) ^ have′ (1sg, knig-) for Russian in (95a), exist′
([have′ (3sg, bayan)]) ^ have′ (3sg, bayan) for Guugu Yimidhirr in (96a), and
exist′ (libro) ^ have′ (babae, libro) for Tagalog in (97a). The Tagalog case is
the most interesting one, because, as noted earlier, there is no obvious way
to relate the possessor to the existential predication, and the constructional
LS supplies the connection. Latrouite and Van Valin (2014) give an analysis of
what are called ‘event existential constructions’ in Tagalog, which, they
argue, are a recent innovation and are related to possessive constructions,
roughly, ‘the woman has a book which was bought’ ) ‘there was an event of
book buying by the woman’. Their RRG account of the transition from
possessive to event existential construction crucially involves both the exist-
ential and the possession components in the LS. Thus, these representations
make explicit what is similar and what is different with respect to the
meaning of comparable constructions across languages.

1.4.3 Semantic Roles

RRG makes use of two types of semantic roles: traditional thematic rela-
tions, such as agent, patient, instrument, recipient, goal, theme, etc., and
semantic macroroles, of which there are only two, actor and undergoer.49
See Kailuweit (Chapter 4) for a detailed discussion of semantic roles.
1.4.3.1 Thematic Relations

RRG follows the approach pioneered by Gruber and Jackendoff of deriving
thematic relations from decomposed lexical representation for verbs. It does
not, however, use the same decomposition system as Gruber and Jackendoff;
rather it uses the Aktionsart-based system presented in Section 1.4.2.1. In the
RRG system only state and activity predicates have arguments, and therefore
different semantic classes of verbs reflect the subclasses of state and activity
predicates. Thematic relations are defined in terms of these classes. The full
list of classes is given in Kailuweit (Chapter 4), so only a few examples will be
given here (see also Van Valin 2005: 55).
(98) a. State predicates

1. State or condition dead′ (x) x ¼ Patient
2. Pure location be-LOC ′ (x, y) x ¼ Location, y ¼ Theme
3. Perception see′ (x, y) x ¼ Perceiver, y ¼ Stimulus
4. Cognition know′ (x, y) x ¼ Cognizer, y ¼ Content
5. Internal sensation feel′ (x, y) x ¼ Experiencer, y ¼ Sensation
6. Possession have′ (x, y) x ¼ Possessor, y ¼ Possessed
b. Activity predicates
1. Unspecified activity do′ (x, Ø) x ¼ Effector
2. Motion do′ (x, [walk′ (x)]) x ¼ Mover50
3. Consumption do′ (x, [eat′ (x, (y))]) x ¼ Consumer, y ¼ Consumed
4. Creation do′ (x, [build′ (x, (y))]) x ¼ Creator, y ¼ Creation
5. Directed perception do′ (x, [see′ (x, (y))]) x ¼ Observer, y ¼ Stimulus
6. Use do′ (x, [use′ (x, y)]) x ¼ User, y ¼ Implement
An important feature of the RRG system is that the basic state and activity
predicates take no more than two arguments, and this entails that three-
place predicates must be composed of more than one basic predicate. The
primary way this is done is through forming causative LSs (see Table 1.4),
and the vast majority of three-place predicates are in fact causative.
The argument structure of a verb is the sum of the arguments of its
constituent state and activity predicates. For example, Bill knows algebra
would be know′ (Bill, algebra). Bill learned algebra would be BECOME know′
(Bill, algebra), with no change in the argument structure. Teach is a three-
place predicate, as in Mr Smith taught Bill algebra, and as predicted, it has a
causative LS, namely, [do′ (Mr Smith, Ø)] CAUSE [BECOME know′ (Bill,
algebra)]. The argument structure of teach is the combined argument struc-
tures of the constituent state and activity predicates, namely effector, cog-
nizer and content. This has important theoretical consequences, as
discussed in Kailuweit (Chapter 4) and Van Valin (2005, §2.4.1).

1.4.3.2 Semantic Macroroles

The two semantic macroroles, actor and undergoer, are key elements in the
linking algorithm. They correspond to the pre-theoretical notions of ‘logical
subject’ and ‘logical object’, or alternatively, to the general notions of
‘agent’ and ‘patient’. The latter are often used in two different senses.
‘Agent’ can be narrowly construed to refer to the intending, wilful, control-
ling and initiating participant, and ‘patient’ can be narrowly interpreted as
the participant undergoing a change of state or condition, in which case a
verb like kill would take an agent and a patient but see or know would not. On
the other hand, there is the general sense of ‘agent’ and ‘patient’, more or
less equivalent to ‘logical subject’ and ‘logical object’, such that kill, see and
know all take agent and patient arguments, because they are transitive verbs.
The narrow sense corresponds to them as thematic relations, and the gen-
eral sense corresponds to them as macroroles.
With respect to morphosyntactic coding, the actor and undergoer argu-
ments are (almost always) direct arguments. In a language like English
which relies on word order to signal who is doing what to whom, it means
they are not marked by adpositions. In languages with case systems like
German and Dyirbal, the macroroles are nominative or accusative (German,
Dyirbal first- and second-person pronouns), or they are ergative or absolutive
(Dyirbal third-person arguments). Macroroles do not appear in the dative,
instrumental or other cases as a default.51 In particular, so-called ‘dative
subjects’ in languages like German and Icelandic are not actors; see Van
Valin (1991, 2018b) for arguments to this effect. The same holds for the
‘dative objects’ of verbs like ‘help’ and ‘thank’ in numerous languages; they
are not undergoers (see Van Valin 2018b: 123, fn. 7). Rather, the dative
arguments are non-macrorole direct core arguments. The major exception
to the claim that macroroles are always direct arguments is the construc-
tionally specified status of the ‘demoted’ actor in a passive construction; it is
always an oblique of some kind: for example English by PP, German vonþdat
PP, Japanese ni(yotte) PP, Russian instrumental case RP, Jakaltek -u PP, Bahasa
Indonesia untuk PP.
Macroroles are related to thematic relations by means of the Actor-
Undergoer Hierarchy (AUH) in (99).
(99) Actor-Undergoer Hierarchy
The actor is the most agent-like argument in the LS, namely the leftmost,
while the undergoer is the most patient-like argument in the LS, namely the
rightmost. There is an interesting asymmetry between actor and undergoer

here: the actor is always the leftmost argument in the LS, whereas the under-
goer as the rightmost argument is, for some verbs in some languages, only the
default situation. This follows from an asymmetry in lexicalization. The
meanings of the macroroles are very general: the actor is the participant
who is responsible for the state of affairs, in the sense that there can’t be an
event of running without a runner, there can’t be a cognitive event without a
cognizer, a perceptual event without a perceiver, etc., whereas the undergoer
is the most affected participant in the state of affairs. Languages lexicalize
responsibility but not necessarily affectedness. As examples of variable affect-
edness, consider the pairs of examples in (100).
(100) a. Marie taught French to the students (but they didn’t learn a word of it).
a′. Marie taught the students French (?but they didn’t learn a word of it).
b. The workmen loaded the hay on the truck. (all of the hay, the truck may or
may not be full)
b′. The workmen loaded the truck with the hay. (the truck is full (preferred),
all of the hay may or may not have been loaded)
c. The company shipped the package to Henry (but he never received it).
c′. The company shipped Henry the package (?but he never received it).
In all of these examples, the undergoer is the direct RP immediately

following the verb. The default selection for undergoer is given in the first
member of each pair: French, the hay and the package would all be the
rightmost argument in the LS. It is possible with most three-place verbs in
English to select the second to rightmost argument as undergoer, as in the
second member of each pair, and there is a semantic contrast: the students
were affected by the teaching in (a′), that is, they learned French; the hay is
the primarily affected participant in (b) but the truck is in (b′); and (c)
doesn’t imply that Henry received the package while (c′) does. How a partici-
pant is affected differs from verb to verb, but undergoer choice can have
semantic consequences.
This leads to a more general point: macrorole status has semantic conse-
quences, and this applies to actors as well as undergoers. The thematic
relation ‘agent’ was not discussed in the previous section, and that is
because a full account of ‘agent’ in RRG terms requires the notion of actor.
In (89) and fn. 37, it was mentioned that in Dowty’s system agent arguments
were indicated by DO. This makes having an agent argument a lexical
property of the verb, but this is problematic in some cases. The verb kill
would seem to be a good candidate for a verb that has an agent argument,
but that is questionable, as the examples in (101) show.
(101) a. The hunter killed the deer.

b. The hunter intentionally killed the deer.
c. The hunter inadvertently killed a large dog.
d. Chronic wasting disease has killed many deer.
The default interpretation of (101a) is that shown in (b), not the one in (c),
but both are possible. If the actor of kill is an agent, understood as the wilful,

intending and controlling instigator of an action, then (101c) is unexpected.

Even more unexpected is (d), which has an inanimate subject. Contrast these
examples with those involving the verb murder.
(102) a. The hunter murdered the game warden.

b. ?The hunter intentionally murdered the game warden.
c. *The hunter inadvertently murdered the game warden.
d. *Falling tree branches have murdered many game wardens.
The picture is very different here. The actor of murder must be an agent as
described above, and consequently (b) is redundant, (c) is a contradiction,
and (d) is nonsensical. This is different from the actor of kill, which may,
but need not be, an agent. Rather, following Holisky (1987) and Van Valin
and Wilkins (1996), the actor of kill is an effector, which can be interpreted
as an agent if it is human and there is no evidence to the contrary.
Accordingly, the default interpretation of (101a) is (101b), as noted above,
in which the adverb intentionally specifies that the actor is agentive. In (c)
the adverb inadvertently constitutes evidence to the contrary and blocks
the agent implicature, which does not apply in (d) due to the non-
human effector.
Three points deserve mention. First, the agent implicature applies only to
effector arguments. The human actor of receive or see is not construed as an
agent. Receive has a non-volitional recipient-actor and contrasts with accept,
which has a volitional recipient-actor. See has a non-volitional perceiver-
actor, but watch and look at, which are activity verbs of directed perception
(see fn. 50), are subject to the agent implicature. Second, agent is fundamen-
tally different from all other thematic relations in that it is always an
overlay on top of another thematic relation, most commonly effector. Third,
languages differ in the extent to which they lexicalize agentivity. English
seems to rely on the agent implicature, as does Tsova Tush (Caucasian;
Holisky 1987), whereas in Japanese the human actor of a transitive verb
must be interpreted as an agent (Hasegawa 1996), and the agent implicature
plays no role.
There is also an implicature with undergoer. The first thing to point out is
that in English the undergoer with an active voice transitive verb is the
direct RP immediately following the nucleus, and this means that Sam is the
undergoer in (103a) and the book is the undergoer in (a′). This does not
correspond to the traditional analysis that claims that the book is the direct
object in both sentences and Sam is likewise the indirect object in both.
Evidence comes from passivization, which has the undergoer as the subject
in English. There are two candidates for the passive form of (a), as in (103b, b′);
there is only one possible passive of (a′), The book was given to Sam by Mary.
(103) a. Mary [A] gave Sam [U] the book.52

a′. Mary [A] gave the book [U] to Sam.
b. Sam [U] was given the book by Mary [A].
b′. ??The book was given Sam [U] by Mary [A].

Hudson (1992) conducted a survey of varieties of English and found that

(103b) is acceptable in all varieties but (b′) is not. Many speakers either
interpret it to mean ‘the book received Sam’, which is nonsensical, or found
it to be ill-formed. These facts support the claim that the first RP is the
undergoer in the so-called ‘double-object construction’. The third arguments
in these examples are non-macrorole (NMR) core arguments.
The implicature with undergoer concerns the interpretation of change of
possession vs. change of location, and is exemplified in (104).
(104) a. Mary gave Sam the book (*but he never received it).
a′. Mary gave the book to Sam (*but he never received it).
b. #Mary gave Boston the book.
b′. #Mary gave the book to Boston.
c. Mary sent Sam the book (?but he never received it).
c′. Mary sent the book to Sam (but he never received it).
d. ?Mary sent Boston the book.
d′. Mary sent the book to Boston.
Give is a change-of-possession verb, while send is a change-of-location verb,

and the issue is the interpretation of a human argument as a possessor
with change-of-location verbs. Just as some verbs lexicalize their highest-
ranking argument as an agent, there are verbs which unequivocally
lexicalize their second highest ranking argument as a possessor. The LS
for give would be [do′ (x, Ø)] CAUSE [INGR have′ (y, z)] (alternatively, [do′
(x, [act.on′ (x, z)])] CAUSE [INGR have′ (y, z)]). The success of the transfer of
the book to Sam cannot be denied, as the impossibility of adding but he
never received it shows. When the y-argument is a location, as in (b, b′), the
result is a semantic anomaly, since locations are not normally construed
as possessors (but see Section 1.4.2.3). The LS for send, by contrast, is [do′
(x, Ø)] CAUSE [INGR be-LOC ′ (. . ., z)] (alternatively, [do′ (x, [act.on′ (x, z)])]
CAUSE [INGR be-LOC ′ (. . ., z)]), with the result state predicate being loca-
tional rather than possessional. This means that Sam in (104c, c′) is a
location rather than a possessor, strictly speaking, but if the change-of-
location is successfully completed, the primary interpretation is that Sam
possesses the book. Undergoer selection is crucial for the interpretation of
these examples. When the RP Sam is coded as a non-macrorole oblique
argument, as in (c′), the change-of-location may or may not have been
successfully concluded. The default interpretation is that it was successful,
but adding but he never received it does not generate a contradiction. If
successful, Sam can be seen as the possessor of the book. If, on the other
hand, the RP Sam is expressed as the undergoer of send, then the strong
implication is that the change-of-location was successful, hence the oddity
of adding but he never received it, and Sam is construed as the possessor of
the book. With a location as the y-argument, as in (d′), and with it coded as
a non-macrorole oblique core argument, it is interpreted as the endpoint
of a change of location only. Interestingly, when the y-argument functions

as the undergoer, as in (d), the possession implicature is so strong that it

can coerce a reading whereby the RP Boston is construed as some kind of
quasi-animate entity, for instance the Boston office of a company, which
can possess things.
These phenomena have been discussed before in the literature (e.g. Pinker
(1989), Rappaport Hovav and Levin (2008), Beavers (2010, 2011), among
others), and the usual solution is to claim that, in RRG terms, the LS of the
verb changes, that is, ‘. . .CAUSE [INGR be-LOC ′ (y, z)] ) . . .CAUSE [INGR have′
(y, z)]’. This doesn’t capture the flexibility of the interpretation, which is
more appropriately viewed as an implicature which can be cancelled or
weakened. The ‘human location as possessor’ implicature is strengthened
when the human location is the undergoer, as in (104c). The general mean-
ing of undergoer, as mentioned earlier, is the most affected argument, and
in sentences like (104c) this leads to the interpretations of the change-of-
location as successful and the new human location as a possessor of the
moved entity. It is worth noting that the constructions in (95)–(97) in Section
1.4.2.3, which involve the constructional construal of locative-type argu-
ments as possessors, only get that interpretation when the locative-type
argument refers to a human.
1.4.4 Transitivity and the Lexicon

The notions of valence and transitivity are distinct, albeit related. The
former concerns the number of arguments a verb or other predicate takes,
as represented in its LS in RRG, while the latter is typically understood as the
number of direct core arguments associated with the predicator in the
nucleus of a core: one RP for intransitive predicates, two for transitive and
three for ditransitive. RRG takes a somewhat different approach to transitiv-
ity from other theories. It distinguishes two types of transitivity: syntactic
transitivity [S-transitivity] and macrorole transitivity [M-transitivity].
S-transitivity is basically the traditional notion characterized above, namely
one direct core argument ¼ S-intransitive, etc. Far more important for the
RRG linking system is M-transitivity, a concept originally proposed in Nar-
asimhan (1998). The relationship between valence and M-transitivity is given
in Table 1.6.
The meteorological verb snow has no semantic arguments and conse-
quently no macrorole arguments; hence it is M-atransitive. The change-of-
state verb faint and the activity verb limp have one semantic argument
and one macrorole argument; hence they are M-intransitive. Consump-
tion and creation activity verbs may occur with or without a second
argument (e.g. drink (coffee), write (poetry)), but either way they take only
one macrorole and are M-intransitive; see Van Valin and LaPolla (1997,
§§3.2.3.3, 4.2) and Van Valin (1990a, 2004) for detailed discussion. It is the
active accomplishment versions of these verbs (e.g. drink a cup of coffee,
write a poem) which have two macrorole arguments and are M-transitive.

Table 1.6 Valence, macrorole number and M-transitivity
Valence Macroroles
snow 0 0 M-atransitive
faint 1 1 M-intransitive
limp 1 1 M-intransitive
drink [ACT] 1 or 2 1 M-intransitive
drink [ACTACC] 2 2 M-transitive
smash 2 2 M-transitive
show 3 2 M-transitive
give 3 2 M-transitive
In general, two-argument activity predicates without a specific incremen-

tal theme argument are M-intransitive, while their active accomplish-
ment counterparts are M-transitive, and one piece of evidence for this is
the oddity of Pizza was eaten by Sam for ten minutes (activity) as compared
with The pizza was eaten by Sam in ten minutes (active accomplishment).
There are two major exceptions to this generalization, namely, activity
verbs of directed perception, such as watch, and of use, such as use; they
are M-transitive as activities. Significantly, they do not have an incremen-
tal theme argument and therefore do not participate in the activity–
active accomplishment alternation.
The remaining verbs (smash, show, give) have either two or three semantic
arguments, but all are M-transitive, because there is no such thing as ‘M-
ditransitive’, since there are only two macroroles; see Van Valin (2004) for
extensive argumentation against the existence of a third macrorole. The
third semantic argument of a three-argument verb is a non-macrorole core
argument, as mentioned earlier, which may be direct (e.g. Max [A] gave/
showed Anna [U] a photo of Sam [NMR]) or oblique (e.g. Max [A] gave/showed a
photo of Sam [U] to Anna [NMR]).
M-transitivity plays the same role in RRG that syntactic subcategorization
information plays in theories that employ it; they both define the relevant
notions of transitivity in the different approaches. Syntactic subcategoriza-
tion properties have long been thought to be highly idiosyncratic, although
Chomsky (1986) famously claimed that c[ategory]-selection should follow
from s[emantic]-selection. M-transitivity is not random, as macrorole assign-
ment for the majority of verbs follows the Default Macrorole Assignment
Principles in (105).
(105) Default Macrorole Assignment Principles

a. Number: the number of macroroles a verb takes is less than or equal to the
number of arguments in its logical structure
1. If a verb has two or more arguments in its LS, it will take two macroroles.
2. If a verb has one argument in its LS, it will take one macrorole.
b. Nature: for verbs which take one macrorole,
1. If the verb has an activity predicate in its LS, the macrorole is actor.
2. If the verb has no activity predicate in its LS, the macrorole is undergoer.

There are two principles here, the first one governing the number of macro-
roles a verb takes based on the number of arguments specified in its LS, while
the second governs which macrorole an M-intransitive predicate takes. The
crucial idea in (105b) is that actor macroroles are tied to the presence of an
activity predicate in the LS of M-intransitive verbs. There is an important
asymmetry between these two principles: there are very few exceptions to
(105b) concerning the identity of the single macrorole, but there are numer-
ous exceptions to (105a), which reflects the fact that the same basic semantic
content can be expressed more than one way. An example from the domain of
possession in English and German is own and besitzen (M-transitive) vs. belong
(to) and gehören (M-intransitive). All of these verbs have basically the same LS,
have′ (x, y), and own and besitzen follow the default principle in (105a). On the
other hand, belong (to) and gehören do not, and this can be indicated by
specifying the macrorole number in the LS, namely, have′ (x, y) [MR 1].
Nothing else needs to be specified; all of the morphosyntactic properties of
the four verbs follow from the linking algorithm, given their LS and macro-
role number, as will be shown in Section 1.6.
(106) a. kill [do′ (x, [act.on′ (x, y)])] CAUSE [INGR dead′ (y)]
b. receive INGR have′ (x, y)
c. own have′ (x, y)
d. belong (to) have′ (x, y) [MR 1]
e. arrive INGR be-at′ (x, y)
f. go do′ (x, [move.away.from.ref.point′ (x)]) ^ PROC covering.path.
distance′ (x, (y)) ^ FIN be-LOC ′ (z, x)
g. seem seem′ (x, y) [MR 0]53
h. see see′ (x, y)
i. watch do′ (x, [see′ (x, y)])
j. show [do′ (w, Ø)] CAUSE [INGR see′ (x, y)]
k. run do′ (x, [run′ (x)])
l. drink do′ (x, [drink′ (x, y)])
m. melt BECOME melt′ (x)
n. afraid feel′ (x, [afraid′ (y)])
The lexicon is a vital component in the RRG system, and it is more than
just a storehouse for words and morphemes. It can be divided into two
components, which may be metaphorically labelled ‘the warehouse’ and
‘the workshop’. The workshop is the location where lexical and derivational
rules apply, and therefore it is where the LS for the semantic representation
is assembled, among other operations. Crucially, the lexical items to fill the
argument positions in the semantic representation are selected at this point.
More will be said about this in Section 1.6.
1.5 Information Structure
The third major representation RRG assigns to a sentence is its information

structure; see Bentley (Chapter 11) for an introduction to the RRG approach

Figure 1.33 The constituent, operator and information structure projections of the clause
to information structure. In Figure 1.1 it is instantiated by the arrow paral-

leling the linking algorithm, which symbolizes the fact that information
structure interacts with the linking algorithm in important ways, which is
discussed by Latrouite and Van Valin (Chapter 12). RRG syntactic representa-
tions have two projections, the constituent projection and the operator
projection, and one of the advantages of a multi-projection representation
is that it is always possible to add more projections.54 Figure 1.33 is
Figure 12.4 from Chapter 12, which adds the information structure projec-
tion; it is explicated in detail in that chapter.
The components of the information structure projection are, first of all,
the information units (IU), which correspond to the amount of information
contained in a simple wh-expression, that is, what, who, when, where (Lam-
brecht 1994). The bracketing around the IU nodes represents the topic–
comment division of the utterance; in this example, he, the topic expres-
sion, refers to the topic referent, and bought a new car constitutes the
comment about the topic referent. Below that is the partition of the utter-
ance into focus and background. The heavy broken black line indicates
the potential focus domain, the syntactic domain in which the focus can
potentially fall, and the triangle represents the AFD, the part of the utter-
ance that is the focus of the assertion in the context in which it occurs. The
part of the utterance that is outside of the AFD is the background. The
default relationship between the two bifurcations of the sentence is that
the topic expression occurs within the background, and the AFD occurs
within the comment. These notions will be explained in detail in Chapter 11,
and in Chapter 12 their role in the linking between semantics and syntax
will be illustrated. It is an important aspect of RRG that the analysis of
grammatical phenomena can refer to one or more of these three projections
as well as the LS and semantic roles. This will be amply illustrated in the
chapters of this book.

1.6 Linking between Syntax and Semantics in Simple Sentences
The heart of the RRG system is the linking algorithm which maps between the
semantic and syntactic representations. It is bidirectional: it maps the seman-
tic representation into the syntactic representation, and it maps from syntax
to semantics as well. This reflects the fact that language users are bidirec-
tional. A speaker goes from a message to be conveyed to the formal packaging
of it which is to be uttered (semantics to syntax), while a hearer analyses the
formally packaged message and gives it an interpretation (syntax to seman-
tics). As discussed in Chapter 19, RRG is not a neurocognitive processing
model but has the potential to form the basis for one.
Specific aspects of the linking algorithm are discussed in depth in differ-
ent chapters (e.g. grammatical relations in Chapter 5, case assignment in
Chapter 7, and argument structure alternations in Chapter 6). The purpose
of this section is to give an overview of the system as a whole. Each direction
of mapping will be presented separately, starting with semantics-to-syntax
linking. The linking algorithm is governed by a very important principle,
the Completeness Constraint.
(107) Completeness Constraint:

All of the arguments explicitly specified in the semantic representation of a
sentence must be realized syntactically in the sentence, and all of the
referring expressions in the syntactic representation of a sentence must be
linked to an argument position in a logical structure in the semantic
representation of the sentence.
The application of this constraint will be exemplified in the linkings

discussed below.
1.6.1 Linking from Semantics to Syntax

The steps of the semantics-to-syntax linking algorithm are given in (108).
(108) The linking algorithm from semantics to syntax

a. Step 1: Construct the semantic representation of the sentence based on the
LS of the main predicator.
b. Step 2: Assign actor and undergoer, following the AUH.
c. Step 3: Determine the morphosyntactic coding of the arguments of the
main predicator.
i. Select the privileged syntactic argument (PSA), following the PSA
Selection Hierarchy (see (111)).
ii. Assign the arguments the appropriate case markers and/or adpositions.
iii. Assign agreement marking to the main verb or auxiliary, as appropriate.
d. Step 4: Select the syntactic templates for the sentence, following the
syntactic template selection principles.
e. Step 5: Assign the nucleus, the arguments and the adjuncts to positions in
the syntactic representation of the sentence.

Steps 1 and 2 take place in the lexicon, specifically in the ‘workshop’. The
central move in step 1 is the selection of the predicator in the nucleus and
its LS. The argument positions in it must be filled by referring expressions,
and the choice is determined by the activation status of the referents of the
referring expressions; see Chapter 11. In addition, any adjunct LSs must also
be incorporated into the representation, along with the relevant operators.
As an example, the situation to be described is one in which a woman draws
a man’s attention to something in a work situation subsequent to an online
meeting. The verbal predicator selected is show, and the referring expres-
sions to fill the argument positions are Linda, the salesman, and what. In
addition, there are two predicative prepositions, after and in, which take
one referring expression each, namely the meeting and the office, respectively;
the other argument position is filled by the LS of show. Finally, the operator
values are determined, specifically past tense and interrogative IF. The result
is the representation in (109).
(109) ⟨IF INT ⟨TNS PAST [be-after′ (meeting, [be-in′ (office, [[do′ (Linda, Ø)] CAUSE
[INGR see′ (salesman, what)]])])]⟩⟩
Step 2 is to assign macroroles, and while Linda must be the actor, the speaker
has a choice with respect to the assignment of undergoer. The default is that
what would be the undergoer, and that choice will be reflected in the
following discussion. The result of Step 2 is (110).
(110) ⟨IF INT ⟨TNS PAST [be-after′ (meeting, [be-in′ (office, [[do′ (A: Linda, Ø)] CAUSE
[INGR see′ (NMR: salesman, U: what)]])])]⟩⟩
It should be noted that this is not a distinct level of representation; the

semantic representation from (109) is merely informationally enriched with
macrorole information.
The remaining steps take place outside of the lexicon. Step 3 concerns the
morphosyntactic coding of the arguments in the LS, and the most important
part of it is PSA selection. (See LaPolla, Chapter 5, for explication of the RRG
theory of grammatical relations.) The PSA, as is clear from the name, is the
most important core argument, and in many cases it corresponds to the
traditional ‘subject’. The most significant difference between them is that
the notion of PSA is construction-specific, that is, it is defined as the privil-
eged argument in a particular construction, whereas the notion of ‘subject’
applies to the language as a whole. Thus, one can talk about the ‘PSA of a
construction’ but not ‘the subject of a construction’. The traditional notion
of ‘subject’ is a generalization across the PSAs of particular constructions in
languages in which the PSAs of the majority of constructions are the same.
(See Watters, Chapter 6, for an analysis of the most important constructions
involving argument alternations.) There are numerous examples of lan-
guages in which different constructions have different PSAs (e.g. Tagalog,
Schachter 1976; Jakaltek, Van Valin 1981), leading to the conclusion that

there is no predominant PSA in the system and therefore no true ‘subject’ in

the traditional sense.
The PSA Selection Hierarchy and principles are stated in (111).
(111) a. Privileged Syntactic Argument Selection Hierarchy:
arg of DO > 1st arg of do′ > 1st arg of pred′ (x, y) > 2nd arg of pred′ (x, y) >
arg of pred′(x)
b. Privileged syntactic argument selection principles:

1. Accusative constructions: Highest-ranking argument in (111a) (default)
2. Ergative constructions: Lowest-ranking argument in (111a) – default
c. Restrictions on PSA in terms of macrorole status:
1. Languages in which only macrorole arguments can be PSA: Croatian, . . .
2. Languages in which non-macrorole direct core arguments can be PSA:
Icelandic, . . .
The hierarchy in (111a) resembles the AUH, except that it is unidirec-

tional: ‘arg of DO’ is the highest-ranking argument, ‘arg of pred′ (x)’ the
lowest. In accusative constructions, the highest-ranking argument is the
(default) choice for PSA. The qualification ‘default’ is in parenthesis,
because in some languages, those that lack voice systems, the highest-
ranking argument is the only possible choice for PSA. In languages with
a voice system, the highest-ranking argument is merely the default choice
of accusative constructions. In ergative constructions, on the other hand,
the lowest-ranking argument is always only the default choice, and there
exist grammatical means to allow the highest-ranking argument to be the
PSA, normally an antipassive voice. (See Chapters 5 and 6 for discussion of
voice in RRG.)
A very important consideration in PSA selection is given in (111c), namely,
whether only macrorole arguments can be selected as PSA, or whether non-
macrorole arguments can function as the PSA in a syntactic construction.
This issue is standardly characterized as ‘are so-called dative subjects real
subjects?’, because dative is the default case for non-macrorole direct core
arguments (see Nakamura 2021, this volume, Chapter 7; Van Valin 1991,
2018b). Both German and Icelandic have verbs which appear to have a dative
‘subject’, as shown in (112).
(112) a. Mir schmeck-t das Brot. German

1sg.dat taste-3sg.pres the.nom.neut.sg bread
‘I like the bread,’ lit. ‘The bread tastes [good] to me.’
b. Þeimi lík-ar matur-in-n Icelandic
3pl.dat like-3sg.pres food-def-m.sg.nom
‘They like the food,’ lit. ‘The food likes to them’.
In both languages, the default word order is for the dative RP to be first in
the core, and the finite verb agrees with the nominative RP. If we put these
clauses into a conjunction reduction construction, as in (113), the different
status of the dative RPs is clear (Rögnvaldsson 1982; Zaenen et al. 1985).55

(113) a. Þeimi lík-ar matur-in-n og proi borð-a mikið.

3pl.dat like-3sg.pres food-def-m.sg.nom and eat-3pl.pres much
‘They like the food and eat much.’
b. *Ihneni schmeck-t das Brot und proi essen viel.
3pl.dat taste-3sg.pres the.nom bread and eat much
‘*They like the bread and eat a lot.’
b′. Ihneni schmeck-t das Brot und siei essen viel.
3pl.dat taste-3sg.pres the.nom bread and they.nom eat much
‘They like the bread and they eat a lot.’
In a conjunction reduction construction in a right-branching language,

there are two PSAs: the missing argument in the second clause (the pivot)
and the argument in the first clause which supplies the interpretation of the
missing argument (the controller). In Icelandic, the dative argument in the
first clause can serve as the controller of the nominative pivot in the second
clause, as the (113a) example shows, but this is not the case in German, which
does not permit dative controllers in this construction; consequently, there
must be a nominative pronoun in the second clause, as in (113b′). In this
construction, Icelandic permits non-macrorole direct core arguments in the
dative case to function as a PSA, whereas German does not; it restricts PSA-hood
to macroroles only. (It is also possible in this construction for the controller to
be nominative and the pivot to be the dative argument of the verb in Icelandic
but not in German.) In both languages, however, person agreement on a finite
verb or auxiliary is restricted to macrorole arguments, namely the nominative
RP. This highlights the construction-specific nature of grammatical relations. In
German, the controller of person agreement on finite verbs and auxiliaries is
the same as the controller in the conjunction reduction construction, namely
the highest-ranking macrorole core argument. In Icelandic, on the other hand,
the controller in the conjunction reduction construction is the highest-ranking
direct core argument, regardless of whether it is a macrorole or not, which is
different from the controller of person agreement on the finite verb or auxil-
iary, which is the highest-ranking macrorole core argument.56 Accordingly,
German is consistent in having the same PSA for more than one construction,
while Icelandic is not consistent, with different constructions having
different PSAs.
The other two parts of step 3 in (108c) concern case and preposition
assignment, on the one hand, and agreement, on the other, both of which
were touched on in the above discussion. The RRG approach to case assign-
ment is discussed in Nakamura (Chapter 7), and preposition assignment is
presented in Ibáñez Cerda (Chapter 10). The basic case assignment rules for
direct core arguments are given in (114); they refer to case assigned by verbs,
not case assigned by adpositions.
(114) a. Accusative constructions:

1. Assign nominative case to the highest-ranking macrorole.
2. Assign accusative case to the other macrorole.
3. Assign dative case to non-macrorole arguments (default).

b. Ergative constructions
1. Assign absolutive case to the lowest-ranking macrorole.
2. Assign ergative case to the other macrorole.
3. Assign dative case to non-macrorole arguments (default).
When there is only one macrorole in a core, it counts as the highest ranking
for (114a) or the lowest ranking for (114b). The English preposition assign-
ment rule that is relevant to the example in (109) is given in (115).
(115) To-assignment rule: Assign to to the non-macrorole a argument in the LS configuration

BECOME/INGR/FIN predicate′ (a, b)
This says that when a two-place state predicate is in a change-of-state LS and

the first argument is not a macrorole, it is marked by to. Finally, the rule for
triggering person agreement on a finite verb or auxiliary is the same in
English as the rule in German and Icelandic discussed above.
Accordingly, the output of step 3 of the linking algorithm is given
in (116).
(116) ⟨IF INT ⟨TNS PAST [be-after′ (meeting [ACC], [be-in′ (office [ACC], [[do′ (A: PSA
[NOM] Linda, Ø)] CAUSE [INGR see′ (NMR: to [ACC] salesman, U: [ACC] what)]])])]⟩⟩
{ACT: 3sg}
A number of points need to be clarified. First, nominative case and accusa-

tive case in English are vacuous on full RPs and realized only on pronouns.
Second, English prepositions assign accusative case to their objects, vacu-
ously in this example. Third, ‘ACT: 3sg’ under CAUSE indicates ‘active voice’
and the person and number of the PSA. Fourth, the to-rule in (115) comes
into play, because the salesman is the non-macrorole first argument of
. . . INGR see′ (y, z) . . . Fifth, as with the output of step 2, this is not a new
level of representation but an informationally enriched version of the ori-
ginal representation in (109), now containing both semantic and morpho-
syntactic information.
The next step is the selection of the syntactic templates which constitute
the syntactic representation of the sentence. Syntactic templates were intro-
duced in Section 1.2.6, and some of the templates used in English are given
in Figure 1.20. The selection of the syntactic templates is based on the
output from step 3. There are five RPs: meeting, office, salesman (common
sortal nouns), Linda (proper name), and what (interrogative pronoun). These
require three common noun templates, one proper noun template, and one
pronoun template. There are three PPs, two predicative and one non-
predicative, and the appropriate templates are given in Figure 1.20.
The core template is the most important template, and the selection
principles for it are given in (117).
(117) a. Syntactic template selection principle:

The number of syntactic slots for arguments within the core is equal to the
number of distinct specified argument positions in the semantic
representation of the core.

b. Language-specific qualifications of the principle in (a):

1. All cores in the language have a minimum syntactic valence of 1.
2. Argument-modulation voice constructions reduce the number of core
slots by 1.
3. The occurrence of a syntactic argument in the pre- or post-core slot
reduces the number of core slots by 1 [may override (1) above].
The principle in (117a) is common sense: the number of specified argu-

ments in the LS of the predicator corresponds to the number of slots for
arguments required in the core template. There is also a slot required for the
nucleus. This principle reflects the default situation, and there are language-
specific qualifications in (117b). The first is exemplified by languages which
require dummy ‘subjects’ for argument-less meteorological verbs (e.g. English
it is raining, French il pleut, German es regnet). The second concerns voice
constructions in which the default choice for PSA is either omitted (unspeci-
fied) or appears as a peripheral adjunct. The third applies to languages which
can displace XPs to the PrCS or PoCS. The second and third qualifications can
combine to override the first qualification, for example in Who was arrested by
the police?, who is in the PrCS and by the police is in the core-level periphery,
leaving just the nucleus within the core. In the example under discussion, the
occurrence of an interrogative pronoun (what) as one of the arguments of show
invokes (117b3) and also requires a PrCS template.
One of the options for linking a temporal adjunct is to link it to the PrDP
where it functions as a frame-setting topic. In the example under discussion,
this option will be exercised for the temporal adjunct after the meeting.
Consequently, it will be necessary to have a PrDP template. Finally, the
operators in (109) are interrogative IF and past tense, and this means that
there needs to be an auxiliary do in order to have tense occur core-initially as
a signal of interrogative IF. Thus, the templates needed to give the output of
Step 4 are given in Figure 1.34.
Figure 1.34 Syntactic templates required by the output of Step 3

The assembly of the syntactic representation follows the example in

Figure 1.21 and sets the stage for the final step. Step 5 in (108e) says ‘Assign
the nucleus, arguments and adjuncts to positions in the syntactic represen-
tation of the sentence’, but this a major simplification; the full step is given
in (118).
(118) Step 5. Assign the nucleus, arguments and adjuncts to positions in the
syntactic representation of the sentence.
a. Assign the nucleus to the appropriate position in the core.
b. Assign the [WH] arguments to the appropriate positions in the clause.
c. If there is a [þWH] XP,
1. assign it to the normal position of a non-wh-XP with the same function,
or
2. assign it to the pre-core or post-core slot, or
3. assign it to a position within the potential focus domain of the clause
(default ¼ the unmarked focus position).
d. A non-wh-XP may be assigned to the pre-core or post-core slot, subject to
information structure restrictions (optional).
e. Assign the argument(s) of LS(s) other than that of the predicator in the
nucleus to
1. the core- or clause level periphery (default), or
2. the pre-core or post-core slot, or
3. the pre-detached or post-detached position.
The formulations in (118a, b) are intentionally vague and for good reason:
every language does it differently. While there are well-known cross-linguis-
tic tendencies with respect to word (or phrase) order, languages differ in the
very fine-grained constraints on word order, even if they follow these ten-
dencies with respect to larger-grained phenomena. As mentioned earlier, it
is possible to include the nucleus in the English core templates because its
position is virtually always the same: after the PSA and before the other core
arguments and phrasal adjuncts. This is not possible in languages like
German, Croatian and Dyirbal, in which the nucleus has no fixed position,
and in these three languages the linearization constraints on its position in
the core are different. With respect to the linearization of non-wh argument
expressions, the constraints on their order vary dramatically across lan-
guages, as a brief comparison of, for example, English, Croatian, Dyirbal
and German would show.
The situation regarding wh-expressions is considerably simpler: there are
basically three possibilities. First, they could occur in situ, that is, the normal
position of a non-wh-expression with the same function. Second, they could
appear in the PrCS or PoCS, as in English and Dhivehi (see (6)), and third,
they could appear in the unmarked focus position in the clause (see Chap-
ter 11), as in Turkish and Basque. It should be noted that occurrence of a
semantic argument in the PrCS or PoCS does not involve movement, as the
wh-expression is linked directly from the semantic representation to the
PrCS or PoCS; furthermore, there is no co-indexed empty RP or

Figure 1.35 The output of the semantics-to-syntax linking for (109)
unpronounced copy of it in the core, as the tree structure in Figure 1.6

clearly shows, following (117b3). As mentioned before, appearance in the
PrCS or PoCS is not restricted to wh-expressions, as the statements in (118d)
and (118e2) make clear. Direct linking to the PrDP or PoDP is possible only
for adjuncts, because non-wh syntactic arguments in a detached position
normally require a resumptive pronoun in the clause.
The output of step 5 is the finished sentence, which is given in Figure 1.35.
A condensed representation of the linking is given in Figure 1.36. The
operator projection is omitted, and the added information, namely macro-
role status, PSA-hood, case, etc., is represented above the argument to which
these properties are added for clarity of exposition. The numbers refer to the
steps in the linking algorithm.
1.6.2 Linking from Syntax to Semantics

The essence of the syntax-to-semantics linking algorithm can be summed up
in the following steps.
(119) Linking from syntax to semantics (summary)

a. The parser outputs a labelled tree structure.
b. The first step is to derive as much information from the overt
morphosyntactic features of the clause: case marking/word order, the
voice of the verb, adpositions.
c. The second step is to retrieve the LS of the verb from the lexicon and assign
macroroles where possible.

Figure 1.36 Summary of the linking from semantics to syntax for (109)
d. The information from these steps should link everything in the core to the
argument positions in the LS; if there is an element in the PrCS, it will be
linked last, to the remaining unlinked argument position in the LS. If
there are peripheral adjuncts, retrieve the LS of the predicative adposition
and add this LS to the semantic representation.
The linking presupposes that the parser has output a labelled syntactic
tree. This is an idealization that is appropriate for an abstract grammar, but
it does not reflect the nature of real-time sentence processing, in which
interpretation is incremental and begins with the recognition of the first
word of the sentence and does not wait for the entire sentence to be uttered
before starting the interpretive process. (See Chapter 19 for a proposal which
attempts to account for incremental interpretation in an RRG-based sen-
tence processing system.)
The first step is to glean all information from the morphosyntax of the
sentence. In a language with a voice opposition, the voice of the predicator
in the nucleus is crucial, since it signals the function of the PSA, in English,
the first RP in the core, and in German, the nominative RP. With intransi-
tive predicators, on the other hand, voice and the PSA Selection Hierarchy
are irrelevant, because there is only one direct core argument, which can
be either actor or undergoer, depending on the semantics of the predica-
tor. In addition, there is a wh-RP in the PrCS; it carries no clues as to its
function in the sentence in English, as in principle it could be the PSA (e.g.
What upset Mary?), a non-PSA undergoer (e.g. What did Bill buy?), a non-
macrorole direct core argument (e.g. What did Sally show Linda?), or an
oblique core argument (e.g. What did they load the truck with?). In addition,
there are three PPs, one in the core, one in the core-level periphery, and
one in the PrDP.

The second step is to call up the LS of the predicator in the nucleus, in this
case, show, and assign macroroles. The x-argument is assigned actor, but it is
not possible to assign undergoer, due to the fact that either y or z could be
the undergoer with this verb.
The final step is to reconcile the results of the first two steps in order to
satisfy the Completeness Constraint in (107). There is one direct connection:
the RP Linda is the actor, the actor is the x-argument in the LS of show, and
therefore Linda is the x-argument. The other RP in the core, the salesman, is
the object of the non-predicative preposition to and must be linked to either
the y- or the z-argument in the LS. The key here is the preposition assign-
ment rule in (115), which states ‘Assign to to the non-macrorole a argument
in the LS configuration BECOME/INGR/FIN predicate′ (a, b)’. In terms of the
LS for show, the y-argument is the one corresponding to the ‘non-macrorole a
argument in the LS configuration INGR predicate′ (a, b)’ , and accordingly
the RP the salesman must be linked to the y-argument, as shown in
Figure 1.47. That leaves one unlinked RP argument in the PrCS, what, and
one unlinked argument position in the LS, namely the z-argument. In order
to satisfy the Completeness Constraint, the wh-RP must be linked to the z-
argument, and this yields the correct interpretation of it.
The Completeness Constraint is not yet fully satisfied, however, as there
are two unlinked PPs in the sentence. These are adjunct predicative PPs
which license their arguments and must be incorporated into the semantic
representation of the sentence. Since the prepositions are predicative, they
have lexical entries in the lexicon, and accordingly it is necessary to go back
to the lexicon and retrieve their LSs. The question arises as to the order of
composition of the two prepositional predicates. The locative PP is inside the
clause in the core-level periphery, while the temporal PP is a frame-setting
topic outside of the clause in the PrDP. Consequently, the locative LS is
added, taking the LS of show as one of its arguments, and then the temporal
LS is added, taking the locative LS as one of its arguments. When the
operators are added (not shown), the result is the semantic representation
in (116).
1.6.3 Representing General Principles vs. Language-Specific

Grammatical Information
The linking algorithms as presented in (108) and (119) contain both cross-
linguistically valid general principles as well as language-specific gram-
matical rules, constraints and principles. For example, the syntactic tem-
plate selection principles in (117) have one very general principle in (117a)
and three language-specific qualifications in (117b). It is reasonable to
include these exceptions, since they are quite common, but on what basis
can one motivate their applicability to a given language? This can best be
achieved by having language-specific representations of morphosyntactic
information, and in RRG these are known as constructional schemas

(CSs). As an example, consider the German ‘plain’ passive construction,

(120) a. Hans ha-t den Rechner kaputtgemacht.

have-3sg.pres the.acc.sg computer broken.made
‘Hans broke/ruined/destroyed the computer.’
b. Der Rechner ist von Hans kaputtgemacht worden.
the.nom.sg computer be.3sg.pres from broken.made pass.aux
‘The computer was broken/ruined/destroyed by Hans.’
In order to formulate a CS for the construction in (120b), it is necessary to

introduce briefly the RRG theory of voice (see also Watters, Chapter 6; Wu,
Chapter 26). RRG posits two abstract voice constructions: PSA-modulation
voice, which links an argument other than the default choice to be the PSA,
and argument-modulation voice, which gives non-canonical treatment
to the default choice for PSA. These characterizations are independent of the
passive vs. antipassive opposition, since canonical passive and antipassive
constructions involve both. Crucially, however, they need not co-occur; it is
not uncommon for a passive construction to involve only argument modu-
lation, for example Ute, Sama, impersonal passives of intransitive verbs in
German and Dutch, or for an antipassive construction to have only PSA-
modulation, for example Jakaltek; see Van Valin (1981, 2005: 115–120, esp.
fn. 16).
The CS for the ‘plain’ German passive with M-transitive verbs is given in
Figure 1.37; it shares a number of features with the ‘impersonal’ passive
with M-intransitive verbs, and there is also the ‘bekommen-passive’, which has
CONSTRUCTION: German ‘plain’ passive with M-transitive verbs
SYNTAX:
PSA: (111b1, c1); Variable [±pragmatic influence]
Linking: PSA-modulation, Argument-modulation (core-level peripheral PP, Ø)
Template(s): (117b2)
MORPHOLOGY:
Verb: Past participle
AUX: werden (PERF: worden)
P: von
SEMANTICS:
PSA is not instigator of state of affairs but is most affected participant (default)
PRAGMATICS:
Illocutionary force: Not imperative
Focus structure: Predicate focus (default)
Figure 1.37 Constructional schema for German ‘plain’ passive57

quite different properties from the ‘plain’ werden-passive (Van Valin 2003;
Diedrichsen 2004).
The schema specifies the defining syntactic, morphological, semantic and
pragmatic properties of the construction.58 The first line under ‘syntax’,
‘PSA’, states that it is an accusative construction by referring to the PSA
Selection Hierarchy in (111b1) and that the PSA in this construction must be
a macrorole argument, (111c1). ‘Variable [pragmatic influence]’ indicates
that the PSA is not fixed and that the choice of which macrorole functions as
PSA may be influenced by information-structure factors (see LaPolla (Chap-
ter 5), Bentley (Chapter 11), Latrouite and Van Valin (Chapter 12)). The next
line identifies the construction as a voice construction involving both PSA-
and argument-modulation and specifies that argument modulation is
instantiated either by the default choice for PSA occurring in a PP in the
core-level periphery or by it being omitted. This is indicated by inserting a
‘Ø’ in the leftmost argument position of the LS. The third line states that the
exception in (117b2) for argument-modulation voice construction comes
into play in the selection for the syntactic template. Thus, the entries under
‘syntax’ in the CS give the language-specific values of the cross-linguistically
valid principles in the theory.
The next entries under the heading ‘morphology’ give the relevant morpho-
logical properties of the construction. The first entry gives the required form
of the verb, namely the past participle, and the second identifies the auxiliary
used in the construction, namely werden ‘become’, which has an irregular
past-participle form when it is used as a passive auxiliary, worden, which lacks
the normal ge- prefix: Es ist wärmer geworden ‘It has got warmer’ (*Es ist wärmer
worden) vs. *Der Rechner ist von Hans kaputtgemacht geworden (cf. (120b)). The final
part specifies the P in the adjunct PP which contains the actor, the default
choice for PSA. It is von ‘from, of’, and the fact that it takes a dative object
need not be specified here, as that is a general property of von.
The characterization of the meaning of the construction under the
‘semantics’ heading is quite general but appropriate for the construction.
There are two entries under ‘pragmatics’. The first states that this construc-
tion can be used in assertions and questions but not in imperatives, and the
second reflects the fact that the default focus structure is predicate focus,
with a topical PSA and a focal predicate phrase (see Chapter 11), and
accordingly in a passive construction, just as in an active voice form, the
nominative PSA is interpreted as the default topic, although it must be
emphasized that this is only the default interpretation.
CSs thus represent language-specific information, formulated in terms of
the appropriate general, cross-linguistically valid principles. The schemas
play an important role in the linking between form and meaning. In Van
Valin (2005) there is a detailed discussion of the integration of the linking
algorithm with CSs in Sama wh-questions (pp. 132–135, 157). In order to
illustrate the interaction of CSs with the linking algorithm, the schema for
the Sama antipassive is taken from Van Valin (2005) and presented here in

CONSTRUCTION: Sama antipassive
SYNTAX:
Linking: PSA-modulation, argument-modulation: optional if 3p (Ø), obligatory if
~3p] (Ø, core PP)
Template(s): (117b) [only with Ø argument-modulation]
MORPHOLOGY:
Verb: N- + stem
P: ma
SEMANTICS:
PSA is instigator of state of affairs
If predicate is stative, then it shifts to activity in actor voice
If lowest ranking argument is not marked by determiner, then preferentially indefinite
or non-referential
PRAGMATICS:
Illocutionary force: Unconstrained
Figure 1.38 Constructional schema for Sama antipassive
Figure 1.38 is an updated form, along with the relevant data in (121)
(Walton 1986, personal communication). The homorganic nasal prefix N-
which marks actor voice is realized as Nga- when prefixed to a verb begin-
ning with /n/.
(121) a. Ø-Nda’ ku/ d’nda onde’ hi’. Undergoer voice (active)

atv-see 1sg.erg/woman child det
‘I/the woman sees the child.’
a′. Nga-nda’ aku/ d’nda onde’ hi’. Actor voice (antipassive)
antip-see 1sg.nom/woman child det
‘I/the woman is looking at/for the child.’
b. Ø-Nda’ onde’ hi’ aku.
atv-see child det 1sg.nom
‘The child sees me.’
b′. Nga-Nda’ onde’ ma aku.
antip-see child obl1sg.nom
‘The child is looking at/for me.’
c. Say Ø-nda’ d’nda hi’?
who atv-see woman det
‘Who does the woman see?’ (‘*Who is looking for/at the woman?’)
c′. Say Nga-nda’ d’nda hi’?
who antip-see woman det
‘Who is looking at/for the woman?’ (‘*Who does the woman see?’)

The relevant facts about actor and undergoer voices are summarized in
the pairs of examples in (121a, b); the first example is undergoer voice
(active) and the second is actor voice (antipassive). In the (a, a′) examples
the actor is either first person (aku/ku) or third person (d’nda ‘woman’), and
the undergoer is third person (onde’ ‘child’). Sama has case marking only on
pronouns. In the undergoer (active) voice the full RPs are distinguished by
word order, while the pronominal actor occurs in the ergative form. In the
actor (antipassive) voice, the pronominal actor occurs in the nominative
form, signalling that it is the PSA. A striking feature of this construction is
that when the undergoer is third person, it may occur as a direct core
argument.59 If, however, the actor is third person and the undergoer is first
or second person, as in the (b, b′) examples, the situation is very different. In
the undergoer voice, the first-person undergoer occurs in the nominative
form aku, as in (121b), but in the actor voice the first-person undergoer is
marked by the oblique preposition ma plus the nominative pronoun, as in
(121b′). The CS for this voice construction is given in Figure 1.38.
The PSA entry specifies that this is an ergative PSA restricted to macro-
roles, while the linking entry states that the primary function of the con-
struction is PSA-modulation, with argument-modulation optional for third-
person arguments (omission), but obligatory for non-third-person argu-
ments, the options being omission or occurring as an oblique core argument
in a PP as in (121b′). The final entry states that the default syntactic template
selection principle in (117a) applies unless there is optional argument-
modulation, and then only when the choice is omission, since the PP for
non-third persons is in the core, not a periphery. The relevance of this CS to
wh-questions in Sama is clearly illustrated in (121c, c′). Wh-expressions like
say ‘who’ or ay ‘what’ occur in the PrCS and must be the PSA of the clause.
Accordingly, if the verb is in undergoer voice, as in (121c), the wh-expression
must be interpreted as the undergoer, and d’nda ‘woman’ must be inter-
preted as the actor. On the other hand, if the verb is in actor (antipassive)
voice, then the wh-expression is the actor and the postverbal RP d’nda’ must
be interpreted as the undergoer. The CS for wh-questions invokes the anti-
passive CS when the wh-expression is an actor.
Thus, the schema for the Sama antipassive, like the one for the German
passive, combines cross-linguistically motivated theoretical principles with
language-specific information. The interaction of CSs with the linking algo-
rithm will be discussed in more detail in the next section. The issue to be
investigated in the remainder of this section is the range of ways that
language specific-grammatical information can be represented.
CSs are one of the primary devices for representing language-specific gram-
matical information. Some cases of word-order variation lend themselves
readily to a constructional analysis, for instance the occurrence of a phrase
in an extra-core position (PrDP, PrCS, PoCS, PoDP) subject to information-
structural constraints. Some basic word-order patterns also seem amenable
to a constructional analysis, even though a given word order need not be

associated with a specific semantic content, for example in English [CORE RP

[NUC V ] PP ] can instantiate ‘Actor – Activity Verb – Oblique’ (e.g. Sam sang to
Sally. Tom talked to Max. Bill ran to the beach. Fred looked at the parade.) as well as
‘Undergoer – Process Verb – Oblique’ (e.g. The gravy spilled on the table. The
napkin fell on the floor. The ice cream melted onto the counter.). In a language like
English or French, which has relatively rigid word order, the syntactic tem-
plates capture many of the basic word-order patterns. More interesting is the
situation in languages like German, Croatian and Warlpiri, which have very
free phrase order together with one fixed point of reference, namely second
position in the clause. The factors which influence the ordering of the con-
stituents other than the one associated with second position are complex and
involve information structure, phonological weight, inherent lexical content
(‘animacy’), among other things.60
1.6.4 Interaction of Constructional Schemas and the

Linking Algorithm
In the next two sections the interaction of CSs with the semantics-to-syntax
and syntax-to-semantics linking algorithms will be illustrated.
1.6.4.1 Linking from Semantics to Syntax Revisited

The linking algorithm was given in (108) and (118), and (118c) and (e) contain
options which different languages may utilize. These can be replaced by
language-specific CSs, the range of options being a meta-generalization
across the CSs. In principle, CSs can contribute language-particular infor-
mation at each one of the steps in (108). To illustrate this, the linking from
semantics to syntax in two Japanese passive constructions will be illustrated,
namely, the plain passive in (122) and the adversative passive in (123)–(125)
(Kuno 1973; Imai 1998; Toratani 2002), each with its own CS in Figures 1.39
and 1.40.
(122) a. Hanako ga yakuza o korosi-ta.

nom gangster acc kill-pst
‘Hanako killed the gangster.’
b. Yakuza wa Hanako ni(yotte) koros-(r)are-ta.
Gangster top ‘by’ kill-pass-pst
‘The gangster was killed by Hanako.’
c. [do′ (Hanako, [act.on′ (Hanako, yakuza)])] CAUSE [INGR dead′ (yakuza)]
(123) a. Hanako ga nekom-da.

nom become.bedridden-pst
‘Hanako became bedridden.’
a′. INGR bedridden′ (Hanako)
b. Taroo wa Hanako ni nekom-(r)are-ta.
top dat become.bedridden-pass-pst
‘Taro was affected by Hanako’s becoming bedridden.’
b′. feel.affected′ (Taroo, [INGR bedridden′ (Hanako)])

CONSTRUCTION: Japanese ‘plain’ passive
SYNTAX:
Linking: PSA-modulation, argument-modulation: (Ø, core-peripheral PP)
Template(s): (117b)
MORPHOLOGY:
Verb: Stem + (r)are
P: ni(yotte)
SEMANTICS:
PSA is not the instigator of state of affairs but undergoes the action of the verb.
PRAGMATICS:
Figure 1.39 Constructional schema for Japanese plain passive
CONSTRUCTION: Japanese adversative passive
SYNTAX:
PSA: (111b1, c2)
Linking: Highest ranking argument of stem verb —> NMR
Template(s): (117a)
MORPHOLOGY:
Verb: Stem [renyookee ‘infinitive’ form] + (r)are
Case: Assignment follows Imai (1998)
SEMANTICS:
-(r)are adds feel.affected′ (x, [y]) [MR Ø] to LS of stem verb y
PSA feels negatively affected by state of affairs
PRAGMATICS:
Figure 1.40 Constructional schema for Japanese adversative passive
(124) a. Hanako ga odot-ta.

nom danced-pst
‘Hanako danced.’
a′. do′ (Hanako, [dance′ (Hanako)])
b. Taroo wa Hanako ni odor-are-ta.
top dat dance-pass-pst
‘Taro was affected by Hanako’s dancing.’
b′. feel.affected′ (Taroo, [do′ (Hanako, [dance′ (Hanako)])])

(125) a. Hanako ga piza o tabe-ta.

nom pizza acc eat-pst
‘Hanako ate the pizza.’
a′. do′ (Hanako, [eat′ (Hanako, piza)])
b. Taroo ga Hanako ni piza o tabe-rare-ta.
nom dat pizza acc eat-pass-pst
‘Taroo was affected by Hanako’s having eaten the pizza.’
b′. feel.affected′ (Taroo, [do′ (Hanako, [eat′ (Hanako, piza)])])61
Despite the superficial similarities of the two constructions, their CSs

show clearly how different they are. Both constructions appear to be
marked by the same morpheme, -(r)are, but in fact -(r)are has different
functions in the two constructions. The plain passive involves a relinking
of the actor and undergoer in (122a), with the undergoer appearing as the
PSA and the actor as an optional adjunct in the core-level periphery for
phrasal adjuncts as in (122b); the LS of the verb is not affected, as both
(122a) and (b) have the same LS, given in (c). In the adversative passive, on
the other hand, the addition of -(r)are changes the semantic representation
of the core by adding a predicate to the LS of the verb (feel.affected′ (x, [. . .])
[MR Ø]) and licensing an additional argument, and therefore it is a valence-
increasing construction. This is highly unusual for a voice construction and
is more akin to what is found in an applicative construction. There are no
‘active’ forms with the same arguments as the adversative passive, and this
raises the question of whether this construction should even be labelled a
‘passive’ construction. Another apparent similarity between the two con-
structions is that the plain voice PSA appears to be marked by the same
morpheme, ni, but in fact in (122b) ni is a postposition and can have the
alternative form niyotte, whereas in (123)–(125) the ni is the dative case
marker and niyotte is not possible in this construction. Adversative -(r)are
can occur with both transitive and intransitive verbs, and with intransitive
verbs which have an undergoer argument, as in (123), as well as to those
which take an actor, as in (124). The plain passive, on the other hand, is
restricted to transitive verbs.
Macrorole assignment in the adversative forms is different from that in
(122). The highest-ranking argument of the base verb is not assigned a
macrorole but rather is a non-macrorole argument; this predicts that it
should receive dative case, which is correct. The macrorole status of the
argument added by -(r)are is an interesting issue. To begin with, Japanese
allows non-macrorole arguments to be the PSA of some constructions, for
instance as the antecedent (controller) of a reflexive (Kuno 1973), and there
are cases in which a non-macrorole can appear as the nominative PSA in a
passive (Imai 1998); moreover, since it is the leftmost argument in the LS,
regardless of its macrorole status, it will be the PSA according to the PSA
selection principles in (111b1, c2), which is specified in the CS. Thus there is
no reason to assign it a macrorole to account for its selection as the PSA,
hence the ‘[MR Ø]’ notation in Figure 1.40 (Toratani 2002); this applies only

to the argument licensed by -(r)are, since, with a transitive verb like taberu
‘eat’ in (125), the accusative RP is an undergoer.62 Because the non-macrorole
argument introduced by -(r)are is the PSA, it does not take dative case,
following the more complex set of case assignment rules proposed for
Japanese in Imai (1998) instead of the ones given in (114a).
The steps in which the CS in Figure 1.39 plays a role in the linking from
semantics to syntax for (122b) are summarized in (126); the steps were given
in (108).
(126) Linking in (122b)

a. Step 1: LS formation. Not involved.
b. Step 2: Macrorole assignment. Not involved.
c. Step 3: Morphosyntactic coding. Involved in PSA modulation, argument-
modulation (if it involves omission of actor, then the actor argument in
the LS will be blocked from occurring overtly, or if the realization of the
actor is as a core-level adjunct, it is marked by P ni(yotte)), adds suffix -(r)are
to verb stem.
d. Step 4: Syntactic template selection. Choices affected by argument-
modulation in step 3.
e. Step 5: Assignment of elements to syntactic trees. Actor PP assigned to
core-level periphery for phrasal adjuncts, undergoer RP normally in initial
position.
The situation is very different in (123)–(125).
(127) Linking in (123)–(125)

a. Step 1: LS formation. Embed the LS for the verb into the second argument
position in the LS for -(r)are.
b. Step 2: Macrorole assignment. The argument of -(r)are and the highest-
ranking argument of the embedded LS are not assigned a macrorole.
c. Step 3: Morphosyntactic coding. Involved in case assignment.
d. Step 4: Syntactic template selection. (117a) is based on derived LS.
e. Step 5: Assignment of elements to syntactic trees. Dative NMR is assigned
to the core, the argument of -(r)are is in initial position.
As noted in Section 1.6.1, steps 1 and 2 take place in the lexicon, while
steps 3–5 take place in the syntax, outside of the lexicon. In the plain passive
in (122), both active and passive versions of the sentence have the same LS,
given in (122c), and accordingly it does not involve steps 1 and 2. By contrast,
steps 1 and 2 are crucial for the adversative construction, as the combining
of the LSs takes place at step 1 and macrorole assignment (or the lack
thereof ) at step 2. The remaining three steps in (127) reflect the conse-
quences of steps 1 and 2. In (126), on the other hand, all of the relevant
steps are outside of the lexicon, namely, steps 3 through 5. The adversative
construction is, therefore, considered to be a lexical construction, while
the plain passive is a syntactic construction. This distinction has
important consequences for certain phenomena, for example reflexive bind-
ing (see Van Valin 2005, §5.2).

1.6.4.2 Linking from Syntax to Semantics Revisited

The algorithm for linking from syntax to semantics in simple sentences
given in Van Valin (2005: 149–150) includes both cross-linguistically valid
principles and language-specific features, and it has been argued that
language-specific information should be represented in CSs rather than in
the linking algorithm directly. Accordingly, there must be schemas repre-
senting the basic clause structure of a simple sentence for a language, as
well as for the voice oppositions, if any, that a language has. Preliminary
schemas for basic clause structure, displaced wh-questions, and peripheral
adjuncts in English are given in Figures 1.41 to 1.43.
The CS in Figure 1.41 summarizes the basic facts about a simple sentence
with a single core containing a single nucleus. It does not include the PrDP
or PrCS, since they are not part of a basic simple clause.
The CS in Figure 1.42 highlights the striking contrast between the simpli-
city of linking from semantics to syntax in displaced wh-questions in English
and the complexity of linking from syntax to semantics. Regardless of
whether the wh-expression is a core argument or a peripheral adjunct, the
linking from semantics to syntax is straightforward: the wh-expression
occurs in the PrCS.
The CS in Figure 1.43 is concerned with peripheral adjuncts, and there are
two types. The first includes one-place adverbial modifiers, such as tomorrow
in tomorrow′ ([do′ (Sam, [visit′ (Sam, Meredith)])]), Sam visits Meredith tomor-
row, where the LS of the core is the single argument of the adverbial predi-
cate. The second type is predicative PPs, in which the object of the predicative
P is the first argument and the LS of the core is the second, as in Sam visited
CONSTRUCTION: English basic clause structure
SYNTAX:
PSA: (111b1, c2)
Linking: Passive construction (optional), variable undergoer selection (default)
Template(s): (117b1,2); RP, PP templates, see CSs for RP, PP
Order: [S [CL [C RPPSA {AUX …}[N PRED ] (RPU) (RP/PP)] Adjunct(s)]]
MORPHOLOGY:
Auxiliary: [±past] PSA agreement; [-tense] bare infinitive, participle[±past]
Verb: [±past] PSA agreement; [-tense] bare infinitive, participle[±past]
RP Case: [+PRO] NOM, ACC, GEN; [-PRO] GEN
SEMANTICS:
LS of PRED in nucleus plus arguments and adjuncts
PRAGMATICS:
Illocutionary force: Unspecified
Figure 1.41 Constructional schema for basic clause structure in English

CONSTRUCTION: English displaced wh-questions
SYNTAX:
PSA: none
Linking [semantics to syntax]: WH-XP in PrCS
[syntax to semantics]:
Arguments: After all of the XPs in the core have been linked, link the
WH-XP to the remaining unlinked argument position in the semantic
Adjuncts: If there are no unlinked argument positions in the semantic
representation of the core, treat the WH-word like a predicative
preposition and follow the CS for linking adjuncts
Template(s): (117b3), PrCS
MORPHOLOGY:
X[+tense] must be the first element in the core linearly.
SEMANTICS:
Contains an open proposition with variable α , WH-XP = α.
PRAGMATICS:
Illocutionary force: Interrogative
Focus structure: Narrow focus on PrCS
Figure 1.42 Constructional schema for English displaced wh-questions
Meredith in the hospital, be-in′ (hospital, [do′ (Sam, [visit′ (Sam, Meredith)])]), or
Sam visited Meredith after the party, be-after′ (party, [do′ (Sam, [visit′ (Sam,
Meredith)])]). In both examples the first argument is the spatial or temporal
reference point and the second argument is the event or state of affairs being
located in space or time. While they all have a default periphery in which
they occur, they are also subject to displacement and occurrence in the PrCS
and PrDP. In addition, there can be multiple adjuncts, and their ‘scope’-like
relations are constrained by both semantic and pragmatic factors. Many of
the aspects of the CS in Figure 1.43 would be applicable to other languages;
some would require only a superficial reformulation in terms of word order,
for example a language which has postpositions rather than prepositions,
but some would be quite substantial, for example languages which lack a
PrCS and/or a PrDP, and, more significantly, languages such as Dyirbal that
lack adpositions altogether. In languages with rich case systems, very often it
is cases which mark adjuncts rather than adpositions (e.g. Hungarian), and so
it is necessary to recognize that there can be predicative cases as well as
predicative adpositions (see Nakamura 2021 for a discussion of predicative
cases in Russian).
It is now possible to formulate the syntax-to-semantics linking algorithm
in such a way that it contains only cross-linguistically valid steps, the
language-specific information being contained in a language’s CSs. The goal
here is to redo the linking in Section 1.6.2 based on information supplied by
the CSs in Figures 1.41–1.43.

CONSTRUCTION: English peripheral adjuncts
SYNTAX:
PSA: none
Linking [semantics to syntax]:
Assign the argument of an LS other than that of the predicator in the
nucleus to a predicative PP headed by the predicative P, or identify a
one-place adverbial predicate, and instantiate them in
1. a periphery [(81i) = core, (81j) = clause] (default), or
2. the PrCS or the PrDP, subject to information-structural conditions (see
PRAGMATICS).
[syntax to semantics]:
If there are no unlinked argument positions in the semantic represen-
tation of the clause, retrieve the LS for the predicative preposition or
he adverbial from the lexicon; if the predicative element is a one-place
adverb, insert the LS of the core into its argument position, and if it
is a two-place preposition, insert the LS of the core into the second
argument position and insert the object of the P as the first argument.
Template(s): periphery, predicative PP, adverb, PrDP, PrCS
MORPHOLOGY: none
SEMANTICS:
Semantic relations: (81i, j)/(94i, j)
When there are multiple adjuncts, the order of linking reflects (1) clausal
periphery > core periphery, (2) PrDP > PrCS > default periphery, or (3) [PER
AJTTOP > AJTFOC ], where the lower ranking adjunct is linked first.
PRAGMATICS:
Illocutionary force: Unspecified
Focus structure:
Frame-setting topic in PrDP
Contrastive focus/topic in PrCS
Neutral comment, narrow focus in periphery
Figure 1.43 Constructional schema for English peripheral adjuncts
(128) Linking algorithm: syntax to semantics

Step 1: Determine the syntactic and semantic functions of the core arguments,
based on the morphosyntactic properties of the clause. See the CS for the
language’s basic clause structure and the CSs invoked by it.
Step 2: Retrieve from the lexicon the LS of the predicator in the nucleus and
execute the step in (108b), subject to the following proviso: If a language has
verbs which allow variable undergoer assignment, and the core contains
more than one choice for undergoer, do not assign undergoer to an argument
in the LS.
Step 3: Link the arguments identified in step 1 with those identified in step 2.
Step 4: For variable undergoer-selection verbs, determine the linking of the
non-macrorole core argument:
a. If there is a two-place state predicate in the LS and if the non-macrorole
core argument is marked by a locative adposition or dative or a locative-
type case, then link it with the first argument position in the state
predicate in the LS and link the other non-actor core argument (if there is
one) to the second argument position in the state predicate, or

b. If there is a two-place state predicate in the LS and if the non-macrorole

core argument is not marked by a locative adposition or dative or a
locative-type case, then link it with the second argument position in the
state predicate and link the other non-actor core argument (if there is one)
to the first argument position in the state predicate.
c. If neither (a) nor (b) applies, see the relevant CS for the language.
Step 5: If there is a wh-XP in the Pr/PoCS, see the relevant CS for the language.
Step 6: If there is a peripheral adjunct, see the relevant CS for the language.
It might be objected that step 4 contains considerable language-specific

information, but in fact it is surprisingly valid across languages, having
been applied to Croatian, Dyirbal, English, German, Russian, and
many others.
The detailed linking from syntax to semantics for After the meeting, what did
Linda show to the salesman in the office? begins with the parser output, namely,
in Figure 1.44.
The output of step 1 is given in Figure 1.45. It makes crucial reference to
the CS ‘Basic clause structure’ in Figure 1.41, which states that the PSA is
accusative and not restricted to macroroles and that it is the core-initial RP,
that the default position of the undergoer in an M-transitive core is imme-
diately after the nucleus, and that there is the possibility of a passive
Figure 1.44 Parser output in syntax-to-semantics linking
Figure 1.45 Output of step 1 in the syntax-to-semantics linking

Figure 1.46 Output of step 2 in the syntax-to-semantics linking
construction, thereby potentially invoking the passive CS.63 However,

because the voice of the verb is active, the PSA is an actor, and according
to the CS in Figure 1.41, the RP Linda as the first RP in the core, must be the
actor. There is no direct argument immediately following the nucleus,
hence no undergoer in the core, but there is a PP headed by to with the
object the salesman and two adjunct PPs.
Step 2 involves calling up the LS of the predicator in the nucleus, in this
case [do′ (x, Ø)] CAUSE [INGR see′ (y, z)] for show and assign macroroles as in
the comparable step in (108b) in the semantics-to-syntax linking (Figure 1.46).
Accordingly, the x-argument is the actor, but because show is a variable
undergoer verb, it is not possible to assign undergoer.
Step 3 is rather trivial in this case, despite the complexity of the example:
in step 1 Linda is identified as the actor, and in step 2 the x-argument in the
LS is assigned the actor macrorole; ergo, the RP Linda in the syntactic
representation links to the x-argument in the LS of show.
Step 4 deals with the problem of determining the linking of the non-actor
arguments when there is more than one candidate for undergoer. Some
languages mark the choice of undergoer overtly, for instance German schen-
ken ‘to give as a gift’, has the theme (z) as the undergoer, as in (129a), whereas
beschenken ‘to gift someone’, has the recipient (y) as the undergoer, as in
(129b).
(129) a. Der Mann hatder Frau Blumen geschenkt.

the.nom.sg man has the.dat.sg woman flower.acc.pl given.as.a.gift.’
‘The man gave flowers to the woman as a gift.’
b. Der Mann hat die Frau mit Blumen beschenkt.
the.nom.sg man has the.acc.sg woman with flowers.dat.pl gifted.
‘The man gifted the woman with flowers.’
Among its many functions the verb prefix be- serves to signal a non-default
choice for the argument to function as undergoer. Most languages which
have variable undergoer selection are not so straightforward. Rather, the
identity of the argument functioning as undergoer must be deduced from
the way the non-macrorole argument is coded. The following English
examples illustrate this.
(130) a. Mary gave the award to Bill.

a′. Mary gave Bill the award.
b. Mary presented the award to Bill.
b′. Mary presented Bill with the award.
c. [do′ (x, Ø)] CAUSE [INGR have′ (y, z)]

In (130a) the non-macrorole argument is Bill, an oblique core argument,

and therefore step 4 applies to the examples in (130). In (130a) the third
argument is marked by to, which is a locative preposition, and according to
Step 4a this means that the object of to should be linked to the first
argument position (y) in the two-place state predicate in the LS in (130c),
and consequently the other argument, the award, must be linked to the z-
argument in the LS. In (130a′), on the other hand, the third argument, the
award, is not marked by a locative adposition or dative-type case, and
therefore its third argument must be linked to the z-argument, leaving
the recipient to be linked to the y-argument. The same is true in (130b, b′),
which differ from (a, a′) only in the third argument being an oblique core
argument. In (130b′) the third argument is marked by with, which is not a
locative adposition, and therefore the object of with is linked to the z-
argument in the LS in (130c). The outputs of Steps 3 and 4 are given in
Figure 1.47.
There remain three unlinked elements in the syntactic representation,
two predicative PP adjuncts and a wh-expression. They are handled by
Steps 5 (wh-questions) and 6 (adjuncts), each of which refers to a language-
specific CS. Step 5 must be executed in order to link the wh-RP in the PrCS
to the semantic representation. There is only one unlinked argument
position in the LS of the core, the z-argument, and accordingly the RP
what is linked to the z-argument in the LS of show, and this is the correct
linking, because what refers to the content that is shown. Because there
are two adjuncts, they must be composed in a specific order reflecting
their relative ‘scope’. After the meeting occurs in the PrDP and functions as
a frame-setting topic, while in the office is in its default position in the
core-level periphery with no special information-structural function.
Figure 1.47 Output of steps 3 and 4 in the syntax-to-semantics linking

Figure 1.48 The linking from syntax to semantics in terms of the algorithm in (128)
Hence following the CS in Figure 1.43, in the office will be added to

the semantic representation first, followed by after the meeting. Thus the
entire linking is summarized in Figure 1.48, which labels each step of
the detailed linking algorithm in (128) and the relevant CSs invoked
in each.
Step 4c in (128) refers to instances to which Steps 4a, b do not apply. Two
examples of this from Icelandic (Van Valin 1991) and Yaqui (Uto-Aztecan;
Guerrero and Van Valin 2004) are given in (131) and (132).
(131) a. Ég skila-ð-i henni pening-un-um. Icelandic

1sg.nom return-pst-1sg 3f.sg.dat money-def-dat
‘I returned her the money.’
b. Henni va-r skil-að pening-un-um.
3f.sg.dat be.pst-imper return-ptcp money-def-dat
‘She was returned the money.’
b′. *Pening-un-um va-r skil-að henni.
money-def-dat be.pst-imper return-ptcp 3f.sg.dat
‘The money was returned to her.’ [intended meaning]
c. Ég skila-ð-i pening-un-um til hennar.
1sg.nom return-pst-1sg money-def-dat to 3f.sg.gen
‘I returned the money to her.’
c′. Pening-un-um va-r skil-að til hennar.
money-def-dat be.pst-imper return-ptcp to 3f.sg.gen
‘The money was returned to her.’

(132) a. Joan-Ø Peo-ta Ɂuka vaci-ta miika-k. Yaqui

Juan-nom Pedro-acc det.acc corn-acc give-prf
‘Juan gave Pedro the corn.’
b. Peo-Ø Ɂuka vaci-ta miika-wa-k.
Pedro-nom det.acc corn-acc give-pass-prf
‘Pedro was given the corn.’
c. *UɁu vaci-Ø Peo-ta miika-wa-k.
det.nom corn-nom Pedro-acc give-pass-prf
‘The corn was given [to] Pedro.’ [intended meaning]
The issue here is clear: both non-actor arguments are in the same case in the
active voice, and therefore the principles in Steps 4a, b fail to distinguish
between the two arguments in the linking.
The Icelandic verb skila ‘give back, return’ is a three-place predicate whose
non-actor arguments exhibit two different patterns: two dative RPs (recipi-
ent, theme), as in (131a), and a dative RP (theme) and a PP (recipient), as in
(131c). The latter follows the rule in Step 4a, since linking the object of the
preposition til ‘to’ to the y-argument in the LS yields the correct result. This
requires that the preposition til ‘to’ outweigh the dative case on the theme
argument in Step 4a, and consequently this language-specific feature must
be expressed in a CS concerning linking with three-place predicates. The
passive in (131c′) illustrates the fact that the PSA in Icelandic does not have
to be a macrorole; peningunum ‘the money’ has the syntactic privileges of a
nominative PSA, despite not having the coding privileges of one (i.e. nom-
inative case, ability to be an agreement trigger) (see Van Valin 1991).
The pattern with two dative RPs in (131a) renders Steps 4a, b inapplicable,
since the two RPs bear the same case. There is, nevertheless, an important
difference between the two direct core arguments: one is human and the
other is inanimate, and this difference correlates with the way the RPs link
to the LS. The human RP is the recipient (y), while the inanimate RP is the
theme (z). This can be captured in the following step in the Icelandic CS for
three-place predicates.
(133) Animacy Principle:
When the two non-actor RPs in the core are both in the same case, link the
human (animate) RP with the first argument position in the two-place state
predicate in the LS.
This linking rule comes into play in Icelandic only when there are two dative
RP objects in the core; otherwise steps 4a, b in (128) are operative.
A comparison of the Yaqui examples in (132) with the Icelandic examples
in (131a–c) suggests that a principle like (133) is part of the Yaqui linking
algorithm as well, with double-accusative RP objects instead of two dative
RPs. The Yaqui situation is more complex, however, due to sentences involv-
ing valence-increasing constructions, such as derived causative verbs, as

(134) a. UɁu maejto usi-ta mansana-ta yoem-ta miik-tua-k.

det.nom teacher child-acc apple-acc man-acc give-caus-prf
‘The teacher made the child give the man the apple.’
a′. UɁu usi-Ø mansana-ta yoem-ta miik-tua-wa-k.
det.nom child-nom apple-acc man-acc give-caus-pass-prf
‘The child was made to give the man the apple.’
a′′. [do′ (maejto, Ø)] CAUSE [[do′ (usi, Ø)] CAUSE [INGR have′ (yoem, mansana)]]
b. UɁu maejto yoem-ta mansana-ta usi-ta miik-tua-k.
det.nom teacher man-acc apple-acc child-acc give-caus-prf
‘The teacher made the man give the child the apple.’
b′. UɁu yoem-Ø mansana-ta usi-ta miik-tua-wa-k.
det.nom man-nom apple-acc child-acc give-caus-pass-prf
‘The man was made to give the child the apple.’
b′′. [do′ (maejto, Ø)] CAUSE [[do′ (yoem, Ø)] CAUSE [INGR have′ (usi, mansana)]]
These examples have three core arguments in the accusative case, two of
which are human, and only one can be interpreted as the causee and only
one as the recipient. The sentences in (134a, b) are unambiguous: in (a) usi-
‘the child’ can only be interpreted as the causee and yoem- ‘the man’ as the
recipient, and vice versa in (b). Accordingly, the correct linking in this
construction cannot refer simply to case or to animacy.
It was noted in fn. 52 that Yaqui verbs show three different undergoer
linking patterns, termed patterns A, B and C, following Guerrero and Van
Valin (2004). Pattern A is the usual direct–indirect object pattern, exem-
plified by nenka ‘sell’ in (135a), in which the lowest-ranking argument in
the AUH is the undergoer, is in the accusative case, and can be the PSA in
a passive construction, as in (135a′). Pattern B is the primary object
pattern (Dryer 1986), illustrated by miika ‘give’ in (132), in which the
undergoer is the second highest-ranking argument in the AUH, all non-
actor core arguments are in the accusative case, the undergoer normally
immediately follows the nominative actor, and it alone can be the PSA in
a passive. Pattern B not only applies to miika ‘give’ and the other verbs in
this class, but also to derived forms involving valence-increasing construc-
tions, namely causative, applicative, desiderative, reported speech and
propositional attitude. (See Guerrero and Van Valin 2004 for exemplifi-
cation.) Pattern C is defined by variable undergoer selection, alternating
between Patterns A and B, as shown in (135b, c) with chijakta ‘sprinkle’.
(135) a. Joan-Ø Peo-ta-u ju-ka kaba’i-ta nenka-k.

Juan-nom Pedro-acc-dir det-acc horse-acc sell-prf
‘Juan sold the horse to Pedro.’
a′. Ju-Ø kaba’i-Ø Peo-ta-u nanka-wa-k.
det-nom horse-nom Pedro-acc-dir sell-pass-prf
‘The horse was sold to Pedro.’
b. Empo kafe-ta mesa-po chijakta-k
2sg.nom coffee-acc table-loc sprinkle-prf
‘You sprinkled coffee on the table.’

b′. Kafe-Ø mesa-po chijakta-wa-k

coffee-nom table-loc sprinkle-pass-prf
‘Coffee was sprinkled on the table.’
c. Empo kafe-ae mesa-ta chijakta-k
2sg.nom coffee-ins table-acc sprinkle-prf
‘You sprinkled the table with coffee.’
c′. Mesa-Ø kafe-ae chijakta-wa-k
table-nom coffee-ins sprinkle-pass-prf
‘The table was sprinkled with coffee.’
Actor and undergoer selection are best handled in terms of a CS, since
undergoer selection has several language-specific facets. In terms of seman-
tics-to-syntax linking, step 2 in (108b) is the relevant step, and pattern
A follows the default pattern of the AUH. Pattern B, on the other hand, does
not and specifies that the undergoer is the second-highest-ranking argument
in the LS. In a two-argument LS (e.g. jamta ‘break’ [do′ (x, [act.on′ (x, y)])]
CAUSE [INGR broken′ (y)]), ‘lowest ranking’ and ‘second highest raking’ pick
out the same argument, namely y, but in a three-argument LS like miika
‘give’ ([do′ (x, Ø)] CAUSE [INGR have′ (y, z)]) they select different arguments as
undergoer, z in pattern A and y in pattern B. The verbs which exhibit just
pattern A or B do not alternate with respect to undergoer selection, and they
each occur with different case marking. Pattern A verbs like nenka ‘sell’ in
(135a) have an accusative undergoer/theme, (z), while the recipient (y) carries
the postposition -u, making it an oblique core argument. Pattern B verbs like
miika ‘give’ take all of their non-actor direct core arguments in the accusa-
tive case. The undergoer is distinguished by word order: it normally imme-
diately follows the nominative actor, and if it is not immediately after the
NOM RP, it is the first one after it satisfying the selectional restrictions of the
verb. Pattern C verbs like chijakta ‘sprinkle’ show the same case alternation
found in many other languages, namely ACC (U) – LOC (locative oblique core
argument) vs. ACC (U) – INS (instrumental oblique core argument), as in
(135b, c). In all three patterns the undergoer appears in the accusative case;
it is the case marking of the other core arguments that varies across the
three classes of verbs.
The Animacy Principle was introduced in (133) to account for the Icelandic
examples in (131), and based on the superficial similarity between the sen-
tences in (132) and (131) it was suggested that it might be operative in Yaqui,
too. It could be argued that it applies to (132a), for example, but this is also
readily accounted for by Step 4 in (128). More problematic for the Animacy
Principle are the sentences in (136), from Guerrero and Van Valin (2004).
(136) a. Aurelia-Ø Karmen-ta u-ka ili usi-ta bit-tua-k.

Aurelia-nom Carmen-acc det-acc little child-acc see-caus-prf
‘Aurelia showed Carmen the child’, or ‘Aurelia showed the child Carmen.’
b. Tibu-Ø u-ka wikoi-ta juiwa-m64 u’ura-k.
Tibu-nom det-aka rifle-acc bullets-pl take.away-prf
‘Tibu emptied the rifle of bullets.’

In (136a) there are three human RPs, and the sentence is actually ambigu-
ous. It could be argued that the Animacy Principle correctly predicts the
ambiguity of (136a), since it makes it possible for either the RP Karmen or the
RP usi ‘child’ to be interpreted as the first argument in the two-place state
predicate in the LS. With respect to four-place derived predicators, as in
(134), they are not ambiguous, and this can be accounted for by having the
Animacy Principle apply after the actor and undergoer have been identified.
Accordingly, in (134a), after maejto ‘teacher’ is determined to be the actor
and usi- ‘child’ the undergoer, the Animacy Principle can apply to the
remaining two core arguments, and the result is that yoem- ‘person’ is
correctly analysed as the recipient/experiencer, and mansana ‘apple’ is cor-
rectly analysed as the thing given (theme) in the central LS of the events
described in (134).
The sentence in (136b), however, cannot be accounted for by (133), because
all of the non-NOM RPs are inanimate. Hence (133) does not apply, and this is
also the case when the sentence is causativized, as in (137).
(137) Goyo-Ø Tibu-ta u-ka wikoi-ta juiwa-m u’ura-tua-k.

Goyo-nom Tibu-acc det-aka rifle-acc bullets-pl take.away-caus-prf
‘Goyo made Tibu empty the rifle of bullets.’
After determining that Goyo is the causer, actor and PSA and Tibu is the
causee and undergoer, there are still two inanimate arguments left to be
accounted for. All verbs have selectional restrictions constraining the type of
RPs that can function as their arguments, and in RRG they are expressed as
qualia properties (Pustejovsky 1995), both of the verb and of the RPs, such
that the qualia properties of the RPs must be compatible with the selectional
restrictions of the verb’s arguments. This can be formulated as the ‘Qualia
Principle’ as in (138).
(138) Qualia Principle:
In order for an RP to function as an argument of a predicator, its qualia

properties must be compatible with the selection restrictions of that
argument the predicator.
In (136b) and (137), the crucial part of the LS is ‘CAUSE [INGR NOT be-LOC ′
(y, z)]’, where y is the ‘container’ and z are the things in it. With respect
to rifles and bullets, it is clear that rifles can have bullets in them, not
the other way around, and therefore wikoi- ‘rifle’ is the y-argument and
juiwa- ‘bullet’ is the z-argument. Thus the interpretation of (136b) and
(137) derives from the very specific qualia properties of these RPs. By
contrast, invoking the Animacy Principle in (133) to account for (131a)
involves only very general qualia properties, namely animacy. Thus, the
Animacy Principle can be subsumed under the Qualia Principle.
The difference in complexity between what is required for the semantics-
to-syntax linking and the syntax-to-semantics linking is striking. The

former is summarized in (139) and the latter in (140). The steps referred to
in (139) are those in (108), whereas the steps in (140) are those found in
(128). With respect to (139), it is important to keep in mind that patterns
A and B do not involve variable undergoer selection; that is solely a
property of pattern C.
(139) Semantics-to-syntax linking for the three verb patterns

A. Pattern A
1. Step 2: AUH
2. Step 3: Default case assignment rules (DAT is realized by a postposition.)
B. Pattern B
1. Step 2: U is second-highest-ranking argument in LS
2. Step 3: U has ACC case, all non-actor direct core arguments carry
ACC case.
C. Pattern C:
1. Step 2: AUH default, U can be second-highest-ranking argument in LS.
2. Step 3: U is ACC, third core argument is oblique.
D. Derived pattern B verbs (e.g. causatives)
1. Step 2: U is second-highest-ranking argument in LS, regardless of base
pattern
2. Step 3: U bears ACC, all non-actor direct core arguments carry ACC case,
regardless of base pattern.
(140) Syntax-to-semantics linking for the three verb patterns

A. Pattern A
1. Step 1: If active voice, NOM PSA is actor, ACC RP is undergoer, oblique
core argument marked by P. If passive voice, NOM PSA is undergoer,
actor is omitted.
2. Step 2: In LS assign actor and undergoer (lowest ranking argument). If step
1 finds passive voice, replace highest ranking argument in LS with Ø.
3. Step 3: Link arguments identified in steps 1 and 2.
B. Pattern B
1. Step 1: If active voice, NOM PSA is actor, first ACC RP following NOM RP
which satisfies the selectional restrictions of the predicator is U. If
passive voice, NOM PSA is undergoer, actor is omitted.
2. Step 2: In LS assign actor and undergoer (second-highest ranking
argument). If step 1 finds passive voice, replace highest ranking
argument in LS with Ø.
C. Pattern C
1. Step 1: If active voice, NOM PSA is actor, ACC RP is undergoer, oblique
core argument marked by P. If passive voice, NOM PSA is undergoer,
actor is omitted.
2. Step 2: In LS assign actor only. If step 1 finds passive voice, replace
highest ranking argument in LS with Ø.
4. Step 4: Determine the linking of the oblique (non-macrorole) core
argument following Step 4 in (128) .

D. Derived pattern B verbs (e.g. causatives)

1. Step 1: If active voice, NOM PSA is actor, first ACC RP following NOM RP
which satisfies the selectional restrictions of the predicator is U. If
passive voice, NOM PSA is undergoer, actor is omitted.
2. Step 2: In LS assign actor and undergoer (second highest ranking
argument). If step 1 finds passive voice, replace highest ranking
argument in LS with Ø.
4. Step 4: Determine the linking of the remaining arguments, if there are any.
a. Base verb is pattern A: follow Step 4 in (128)
b. Base verb is pattern B: follow the Qualia Principle in (138) .
c. Base verb is pattern C: follow Step 4 in (128)
The point of this discussion of Icelandic and Yaqui is to show what kinds
of principles, both language-specific and cross-linguistically valid, are
needed. The Qualia Principle is a very general principle governing the
relationship between the core arguments and the predicator in verbal cores,
whereas the specification that all non-PSA direct core arguments with pat-
tern B verbs must have accusative case is clearly idiosyncratic to Yaqui. The
information in (138) and (139) would be stored in CSs, under the heading of
‘Linking’ in the syntax section of the CSs.
1.6.4.3 Linking from Semantics to Syntax Revisited Again

Step 5 in the semantics-to-syntax linking algorithm ranged from extremely
vague (‘assign [. . .] to the appropriate positions in the . . .’) to very specific
(e.g. ‘three options for [þWH] CPs’). It is repeated below.
(118) Step 5. Assign the nucleus, arguments and adjuncts to positions in the
a. Assign the nucleus to the appropriate position in the core.
b. Assign the [WH] arguments to the appropriate positions in the clause.
c. If there is a [þWH] XP,
1. assign it to the normal position of a non-wh-XP with the same function, or
2. assign it to the pre-core or post-core slot, or
d. A non-wh-XP may be assigned to the pre-core or post-core slot, subject to
information structure restrictions (optional).
e. Assign the argument(s) of LS(s) other than that of the predicator in the
nucleus to
1. the core- or clause level periphery (default), or
3. the pre-detached or post-detached position.
Rather than specifying the range of cross-linguistic possibilities, it is

preferable to customize the linking algorithm for the language under
investigation, and this can be done by replacing the disjunctive offerings

of (118) with language-specific information from CSs. It may be reformulated

as follows.
(141) Step 5: Assign the nucleus, arguments and adjuncts to positions in the
a. See CS for English basic clause structure [Figure 1.41] (¼a, b)
b. See CS for English displaced wh-questions [Figure 1.42] (¼c)
c. A non-wh-XP may be assigned to the pre-core or post-core slot, subject to
information structure restrictions (optional). [a yet to be formulated CS]
d. See CS for English peripheral adjuncts [Figure 1.43] (¼e)
This better represents what English speakers know about their language and
yet still maintains a balance between cross-linguistically valid linguistic
concepts and principles and the language-specific data that speakers rely
on when they actually use language.
1.6.5 Semantic Motivation and Cross-Linguistic Variation

The semantics-to-syntax linking algorithm in (108) has two phases: steps
1 and 2 take place in the lexicon, while steps 3 through 5 take place outside
of the lexicon in the syntax. There is a striking asymmetry between steps
1 and 2, which will be called ‘the lexical phase’ of the linking, and steps
3 through 5, which will be called the ‘morphosyntactic phase’ with respect
to the amount of cross-linguistic variation each exhibits. In a nutshell, there
is much greater cross-linguistic variation in the morphosyntactic phase of
the linking than in the lexical phase, and from this fact comes the claim
that the more semantically motivated a linguistic phenomenon is,
the less cross-linguistic variation there is, and vice versa. This
can be illustrated for each step.
Step 1 concerns the decomposition of predicating elements, and all lan-
guages make the same fundamental distinctions. Where there is variation, it
concerns the analysis of specific verbs, not variation in terms of the Aktion-
sart categories that mould the system. Languages may vary, for example, as
to whether causative change-of-state predicates are basic or derived from
change-of-state predicates via causativization (Nichols et al. 2004), but
regardless of the direction of derivation, the categories involved are
the same.
Step 2 concerns actor and undergoer assignment. There is more variation
than in step 1, and the main variation is as follows. With respect to actor
selection, there are languages in which the actors of transitive verbs must be
animate (e.g. Lakhota, Japanese), in which the animate actors of transitive
verbs must be interpreted as agentive (e.g. Japanese), and in which the
interpretation of whether an actor is agentive is an implicature for most
verbs, not a lexical property (e.g. Tsova-Tush, English, German). There is
more variation with respect to undergoer selection, and this follows from
the different semantics of actor and undergoer: as noted in Section 1.4.3.2,

actors are responsible for the state of affairs, in the sense that in order to
have a perceptual event there must be a perceiver, for a cognition state of
affairs a cognizer, for an eating event an eater, etc., while undergoer is
associated with affectedness. It appears that languages tend strongly to
lexicalize responsibility but not affectedness, leaving the choice of which
participant is most affected to the speaker in some languages. Languages
vary with respect to whether they allow variable undergoer selection or not.
If they do, there is variation in the number of verbs allowing the variable
selection, and there are differences in terms of the definition of undergoer
in terms of the AUH. As shown earlier, Yaqui has a class of verbs for which
the undergoer is the lowest-ranking argument in the LS of verbs and another
class in which it is the second-highest-ranking argument.
Step 3 involves PSA selection and case marking, and here the cross-
linguistic variation dwarfs that found in steps 1 and 2. The variation in
PSA selection includes syntactically accusative languages without a voice
system, syntactically accusative languages with a productive voice system,
syntactically ergative languages with a productive voice system, mixed
systems with first and second person being syntactically accusative and
third person being syntactically ergative, inverse systems, split-intransitive
systems, Philippine-type languages with multiple voice systems, which can
be syntactically accusative, ergative or symmetrical, and systems which
defy categorization in terms of traditional categories (e.g. Barai, Papua-
New Guinea; Van Valin 2009). Case-marking systems are likewise highly
varied, and they don’t always match the syntactic pattern, for instance it
has long been known that most morphologically ergative languages lack
ergative syntax.
Step 4 is concerned with the selection of the syntactic templates that
constitute the syntactic representation of the sentence, and here too there
is great diversity, given that the templates must embody the syntactic
structures of the full range of human languages. The components of the
LSC fall into two basic groups: those that are semantically motivated (see
Section 1.2.2) and are universal (nucleus, core, clause, peripheries) and
those that are not semantically motivated but rather are pragmatically
motivated and are not universal (PrDP, PrCS, PoCS, PoDP). Step 5, as shown
earlier, incorporates language-specific CSs capturing the full sweep of
linearization options, displacement options, placement of adjuncts, etc.
Thus, the post-lexical steps in the linking algorithm exhibit much greater
cross-linguistic variation than the lexical steps, and this is indicative of
the inverse relationship between semantic motivation and cross-linguistic
variation.
Two further examples will be presented to illustrate this point. The first
concerns the LSC. There are obligatory and optional constituents in it. The
obligatory ones are the nucleus, the core, the clause and the sentence, and
within the core are the arguments of the predicator in the nucleus. As
argued in Section 1.2.2, the fundamental distinctions in the LSC follow from

the nature of human language as a system of communication. Thus the

obligatory aspects of the LSC are semantically motivated. The optional
aspects, the Pr/PoDP and the Pr/PoCS, which are not universal, are not
semantically motivated; rather, they are motivated by discourse-
pragmatic considerations.
The second example is reflexive binding in languages with coreference
reflexives, such as English, German, Icelandic, Japanese, Mandarin. There
are two fundamental features of these constructions that must be
accounted for: (1) the hierarchical relationship between the antecedent
and the reflexive, and (2) the distance between them. The second issue is
uncontroversially syntactic, whereas both syntactic and semantic accounts
of the hierarchical relationship have been proposed. The RRG approach to
coreference reflexives is semantic (see Van Valin and LaPolla 1997, §7.5.2,
Van Valin 2005, §5.3), and the fundamental claim is that the reflexive
anaphor must not be higher on the PSA Selection Hierarchy than the
antecedent. In other words, oversimplifying somewhat, actors bind under-
goers, not the other way around. In most cases involving coreference
reflexives the syntactic c-command approaches and the semantic approach
make the same prediction, but there are a few cases where they make
different predictions, and it turns out that the semantic approach always
makes the correct prediction (see Van Valin 2005, §5.3). There is some
variation in coreference reflexives, but the basic principle, namely that a
higher-ranking argument binds a lower-ranking argument on a semantic
hierarchy, holds up. With respect to the issue of distance between the
antecedent and the reflexive, there is substantial cross-linguistic variation,
ranging from English, which does not allow binding across a core boundary
(e.g. Mary asked Tom to help her/*herself), while Icelandic happily permits this,
to Japanese and Mandarin in which the reflexive anaphor can be bound
across clause boundaries, where the equivalent of Mary said that Tom had
kissed herself would be possible in them. Here again there is substantial
cross-linguistic variation with respect to the aspect of reflexivization which
is not semantically motivated and very little with respect to the aspect
which is.
Thus, the features of human language which are most likely to be univer-
sal, or nearly so, are the ones with the strongest semantic motivation.
1.7 Linking between Syntax and Semantics

in Complex Sentences
The linking algorithms for semantics-to-syntax and syntax-to-semantics

linking in simple sentences were presented in Section 1.6, and the adapta-
tion of them for complex sentences is discussed in detail in Guerrero
(Chapter 14). (See also Van Valin 2005, §7.) Linking in complex RPs, primarily
restrictive relative clauses, is presented in París (Chapter 15) (see also Van

Valin 2012b). Constraints on the formation of wh-questions in complex

sentences are the topic of Shimojo (Chapter 16).
The main aspects of the algorithm that need modification for complex
sentences will be discussed in Section 1.7.1. The final section will examine
two critical areas of the complex sentence linking algorithm that have no
analogue in simple sentences.
1.7.1 Modifications of the Linking Algorithm for Complex Sentences

There are four levels of juncture (nuclear, core, clausal, sentential), and yet
three of the four do not require substantial modifications of the linking
algorithm. Nuclear junctures link like simple clauses with a multi-
argument predicate; there is a single core node and a single clause node,
as in, for example, the Mandarin nuclear juncture in Figure 1.29. Clausal
and sentential junctures, on the other hand, contain constituent clauses or
sentences which link in most instances just like independent clauses or
sentences. An exception regarding clausal junctures will be discussed in the
following section. The major complexity in linking syntax and semantics is
found in non-subordinate core junctures. Core subordination involves either
a clause or a core used as a core argument (e.g. That Bill’s cat ran away shocked
him vs. Bill’s cat’s running away shocked him) or a peripheral ad-subordinate
clause (e.g. Bill’s cat ran away after she got bored with him). The units involved in
these sentences follow the linking algorithm for simple clauses, and they
would link like simple RPs into core positions or like simple adjunct PPs into
the core or clause-level periphery.
Non-subordinate core junctures require that the two cores have an argu-
ment in common, as illustrated in (142) and (143).
(142) a. Max tried to tune the piano. Max is the actor of try and the actor of tune.
b. Sally persuaded Max to tune Max is the undergoer of persuade and the
the piano. actor of tune.
c. Sally promised Max to tune Sally is the actor of promise and of tune.
the piano.
d. Sam brought wine for Wine is the undergoer of bring and drink.
everyone to drink.
(143) a. Sam seems to really like wine. Sam is a syntactic core argument (and
PSA) of the core headed by seem but is a
semantic argument of like.
b. Sally believes Sam to really Sam is a syntactic core argument of the
like wine. core headed by believe but is a semantic
argument of like.
In (142) each clause contains two cores, and there is a single RP which is a
semantic argument of both predicators in the nucleus of each core. In
(143), on the other hand, there is no single RP which is a semantic argu-
ment of each predicate; rather, there is an RP which is a syntactic core
argument, but not a semantic argument, of one core, and a semantic

argument of the predicator in the linked core. Thus, there are two distinct
ways for two cores to have an argument in common: the constructions in
(142) involve what is traditionally known as (obligatory) control, while
those in (143) are traditionally known as raising.65 The argument in
common is technically known as the shared argument in the
construction.
In an obligatory control construction, a participant functions as a seman-
tic argument in each core but is instantiated in only one of the cores. In
sentences like (142a) there is only one core argument in the licensing core,
and it must be the shared argument. In (142b, c) there are two core
arguments in the licensing core, and there has been much discussion over
the past almost sixty years regarding the explanation for the difference
between verbs like persuade (so-called ‘object control’) and promise (so-called
‘subject control’). The solution was given in Foley and Van Valin (1984:
306–311), and it was shown in Van Valin and LaPolla (1997, §9.1.3.1) and
Van Valin (2005, §7.3.1) to be valid for a typologically wide range of
languages. It is semantic: if the licensing verb has causative or jussive
semantics, the undergoer is always the controller, whereas if it does not
have causative or jussive semantics, the actor is the controller. Because this
principle is semantic, it would be expected in light of the discussion in
Section 1.6.5 that it would be universally valid, and as far as is known, this
is the case. This semantic account makes a prediction that syntactic
accounts do not: if a verb can have either jussive or non-jussive semantics,
the choice of controller should change with the semantic change in the
verb. This is illustrated in (144).
(144) a. The teacher asked the rowdy student to leave the classroom.
b. The sick student asked the teacher (to be permitted) to go home.
In (144a) the verb ask has jussive semantics, and the undergoer is the
controller, as predicted. In (144b), on the other hand, it does not have jussive
semantics, as it is a request, and the actor is the controller, as predicted.
Because there is a shared argument in these constructions, the linked core
has one core argument less than if it were a core in a simple sentence, and
therefore the syntactic template selection principles in (117) must be modi-
fied. Unlike the language-specific principles in (117b2, 3), this one is univer-
sal. There is one additional principle affecting non-subordinate core
junctures, which is illustrated here.
(145) a. Bill taught Mary calculus. [3 core arguments]

a′. Bill taught Mary to swim. [2 core arguments þ infinitive]
b. Sam forgot his keys. [2 core arguments]
b′. Sam forgot to lock the door. [1 core argument þ infinitive]
Teach, for example, is a three-place predicate, and one of the arguments

can be an RP, as in (145a), or a proposition, do′ (z, [swim′ (z)]), as in (145a′). In

Section 1.3.2 it was shown that the infinitival cores in these constructions
are not embedded, and therefore they are not core arguments in the licens-
ing core. Accordingly, the core containing the licensing predicate in a non-
subordinate core juncture has one less core argument than when it is a
simple core, as the contrast between (145a) and (a′) shows. The same holds
for two-place predicates like forget in (145b, b′): when it heads a simple core,
as in (145b), it has two core arguments, but when it is the head of the
licensing core in a non-subordinate core juncture, as in (145b′), it has only
one. Thus, the verb in the licensing core in a non-subordinate core juncture
has one less core argument than when it is the head of a simple core with RP
or PP arguments only.
The LSs for (144a′, b′) are given in (146).
(146) a. [do′ (Bill, Ø)] CAUSE [BECOME know′ (Maryi, [do′ (zi, [swim′ (z)])]
b. forget′ (Sami, [[do′ (yi, Ø)] CAUSE [INGR locked′ (door)])])
Both LSs contain an embedded proposition with a lexically unfilled argu-

ment, and it will be manifested as the infinitival core in the core juncture.66
This is a syntax–semantics mismatch: the LS of the linked core is embedded
as an argument of the LS of the licensing core, but in the syntax the infini-
tive is not embedded as a core argument of the licensing core. The fact that
the LS of the infinitival core is an argument of the licensing verb is signifi-
cant, because when the infinitival core is not an argument of the licensing
verb, then the reduction in core arguments does not occur. This is illus-
trated by (142d), Sam brought wine for everyone to drink. The infinitival core is
missing an argument, but the LS of drink is not an argument of bring.
Accordingly, there is no reduction of the number of core arguments in the
core headed by bring.
The revised syntactic template selection principles are given in (147); the
principles unique to complex sentences are in (147b).
(147) Syntactic template selection principles (revised formulation):

a. Basic principle:
The number of syntactic slots for arguments and argument-adjuncts
within the core is equal to the number of distinct specified argument
positions in the semantic representation of the core.
b. Qualifications of the principle in (a):
The number of slots in a core is reduced by 1 if:
1. It is the licensing core in a non-subordinate core juncture in which the
linked core is a semantic argument of the licensing verb.
2. It is the linked core in a non-subordinate core juncture.
c. Language-specific qualifications of the principle in (a):
2. Argument-modulation voice constructions reduce the number of core
slots by 1.
3. The occurrence of a syntactic argument in the pre-/post-core slot
reduces the number of core slots by 1 (may override c1).

There seems to be an interesting asymmetry between (147b1) and (b2),

namely, there seem to be systematic exceptions to the first principle and few
to the second. It will be argued in the following section that the construc-
tion in (143b) is the result of a violation of (147b1), hence its relative rarity in
the world’s languages. Even rarer is a violation of principle (b2), and it is
found primarily in ‘backward control’ phenomena in some Caucasian
languages (e.g. Tsez, Polinsky and Potsdam 2002; Kabardian; Matasović
2008, 2009, 2010). Examples of ‘forward’ (a) and ‘backward’ (b) control in
Tsez are given in (148).
(148) a. Uži-r [t’ek-Ø magazin-yāy yis-a] šuλ’i-s.

boy-dat book-abs store-abl take-inf forget-pst.evid
‘The boy forgot to buy a book from the store.’
[forget′ (užii, [ . . . do′ (yi, [buy′ (yi, . . . ])])
b. Kid-bā ziya-Ø b-išr-a y-oq-si.
girl.II-erg cow.III-abs III-feed-inf II-begin-pst.evid
‘The girl began to feed the cow.’
do′ (xi, [begin′ (xi, [[do′ (kidi, Ø)] CAUSE [do′ (ziya, [eat′ (ziya, Ø]])])
In (148a) the controller and shared argument uži- ‘boy’ is in the licens-
ing core, which is shown by the fact that it carries dative case assigned by
the predicator šuλ’i- ‘forget’, and it is co-indexed with the lexically
unfilled variable in the embedded LS corresponding to the linked core.
This example adheres to (147b1, 2). In (148b), on the other hand, the
situation is very different, as the shared argument and controller is in
the linked core, not the licensing core, as shown by the fact that it carries
ergative case, being the actor of the transitive verb ‘feed’, rather than
absolutive case as the single core argument of the intransitive verb -oq-
‘begin’. As the LS shows, the lexically unfilled variable is in the LS of the
licensing core, and the lexically filled argument is in the LS of the linked
core. Both constructions satisfy the Completeness Constraint, and there is
nothing in the theory which precludes the two possibilities in (148). The
question of why the one option is so much more frequent than the
other remains.
An interesting feature of the (b) example is that the controller in the
linked core triggers noun class agreement on the verb in the licensing core.
This is readily accounted for: ‘begin’ shows noun class agreement with the x-
argument, and the x-argument is co-indexed with kid ‘girl’ (class II) in the
embedded LS, and accordingly it will show class II agreement. The structure
of (148b) is given in Figure 1.49.
This structure violates (147b2), because the linked core has its full comple-
ment of RP arguments, unlike its English translation. It could also be
construed as violating (147b1), because it lacks the core argument in the
licensing core that the principle predicts it should have: -oq- ‘begin’ has two
arguments in its LS, and when it occurs in the licensing core of the juncture,
it should have one core argument.

Figure 1.49 The structure of (148b)
Matasović (2008, 2009, 2010) adduces similar phenomena in Kabardian

and proposes an RRG account based on a distinction between ‘head-first’ vs.
‘dependent-first’ application of grammatical principles. In (148), forward
control in (a) is the result of head-first filling of argument positions with
lexical items and applying case and agreement principles. Backward control
in (b), on the other hand, results from a dependent-first application of these
same operations. Matasović extends this account to the issue of the domain
of case assignment in core junctures. In Van Valin and LaPolla (1997, §9.2.1)
it was argued that some languages have the clause as the domain of case
marking (e.g. English), while others have the core as the domain (e.g.
Icelandic). He shows that the difference between English and Icelandic in
this regard can be explained in terms of the head-first vs. dependent-first
contrast, thereby eliminating the need for two case assignment domains
and extending the explanatory value of this opposition. Framing the issue in
this way opens the door to a possible answer to the question of why one
option is so much more frequent than the other; it has long been recognized
that heads, be they phrasal or sentential, play morphosyntactically signifi-
cant roles in grammar, and it should therefore not be surprising that head-
first rather than dependent-first is the more prevalent orientation for the
application of grammatical rules and principles.
1.7.2 Argument Identity in Complex Sentences

An issue that affects only complex sentence linking is the way argument
identity across different units of the construction is handled.67 There are
a couple of interacting factors. The first is the level of juncture. In a
nuclear juncture involving coordination or cosubordination, the argu-
ments of the component nuclei are ‘pooled’ in a single derived argument
structure, which is often the same as lexical verbs with a similar mean-
ing. For example, the Mandarin nuclear juncture dǎ sı̌ ‘beat to death’ in
(60b) is composed of dǎ ‘beat’ and sı̌ ‘die’; the LS for (60b), as proposed in
Section 1.4.2.3, is [do′ (1sg, [beat′ (1sg, shé)])] cause [INGR dead′ (shé)],

which has three components: dǎ ¼ [do′ (1sg, [beat′ (1sg, shé)])], sı̌ ¼ [INGR
dead′ (shé)], and cause, which represents the constructionally invoked
meaning of causation. The derived LS is very similar to that of a lexical
causative verb, for example smash as in ‘The burglar smashed the
window’ ¼ [do′ (burglar, [act.on′ (burglar, window)])] CAUSE [INGR
smashed′ (window)]. Nuclear junctures with the second predicate being
transitive link like lexical ditransitive verbs, as in the French example in
(149a) (Hyman and Zimmer 1976).
(149) a. Je fer-ai mang-er les gâteux à Jean.

1sg.nom make-1sg.fut eat-inf the.m.pl cakes dat John.
‘I will make John eat the cakes.’ (Literally ‘I will make eat the cakes to John’)
b. Je donner-ai les gâteux à Jean.
1sg.nom give-1sg.fut the.m.pl cakes dat John
‘I will give the cakes to John’.
c. [do′ (1sg, Ø)] CAUSE [do′ (Jean′, [eat′ (Jean, cakes)])] ¼ (a)
d. [do′ (1sg, Ø)] CAUSE [INGR have′ (Jean, cakes)])] ¼ (b)
Actor NMR Undergoer
Macrorole assignment works the same with reference to both LSs, yielding
the prototypical ditransitive linking pattern for both.
The cores in non-subordinate core junctures necessarily ‘share’ a core
argument, as shown in the examples in (142) and (143). The shared argument
was termed the ‘argument in common’ earlier. The constituent projections
of (142a, b) are given in Figures 1.50 and 1.51. (See Section 1.3.2.)
If persuade in Figure 1.51 were replaced by promise, as in (142c), the struc-
ture would not change, because the difference between actor control and
undergoer control is captured by the theory of obligatory control as part of
the linking algorithm. The important point for the discussion of argument
sharing is that there is no empty argument position or zero element in the
Figure 1.50 Core cosubordination in (142a)

Figure 1.51 Core coordination in (142b)
linked core. The essential features of the linking from semantics to syntax
are given in (150).
(150) do′ (Maxi, [try′ (Max, [[do′ (xi, Ø)] CAUSE [INGR tuned′ (piano)]])])
Actori Actori Undergoer
PSA DCA
The x-argument of the embedded LS is lexically unfilled and co-indexed with

the controller in the matrix LS. The co-indexing satisfies the Completeness
Constraint, because the x-argument is realized indirectly via its controller.
See Van Valin (2005, §7.3.1) for detailed discussion.
The ‘raising’ constructions in (143) parallel the obligatory control
constructions in (142a–c): the intransitive ‘raising to subject’ version
with seem is core cosubordination, whereas the transitive ‘raising to
object’ version is core coordination. This is a change from the analysis
in Van Valin (2005, §7.3.2), where it was argued that because deontic
modals are unacceptable in seem constructions (e.g. *Sam must/ought to/
can seem to like wine), the nexus is coordination. But other consider-
ations support a cosubordination analysis. First, in constructions like
(143a) negation in the licensing core can readily trigger an NPI in the
linked core, whereas with the transitive forms the NPI seems awkward.
(151) a. Sam doesn’t/didn’t seem (to Sally) to like (any of ) the wines.
b. Sam doesn’t/didn’t seem (to Sally) to be eating any cookies.
c. Sam doesn’t/didn’t seem (to Sally) to have saved a red cent.
(152) a. Sally doesn’t/didn’t believe Sam to like (?any of ) the wines.

b. Sally doesn’t/didn’t believe Sam to have eaten ?any/the cookies.
c. ?Sally doesn’t/didn’t believe Sam to have saved a red cent.
Second, Bohnemeyer and Van Valin (2017) argue that core cosubordination
constructions have the macro-event property and therefore they cannot take

multiple temporal positional modifiers, while core coordination construc-

tions allow them and therefore lack the MEP.
(153) a. *This morning Sam didn’t seem to be eating pizza in the afternoon.
b. This morning Sally believed Sam to be eating pizza for lunch at noon.
Taken together, the facts in (151)–(153) suggest that constructions like

(143a) are core cosubordination, while those in (143b) are core coordination.
The constituent projections of these two examples are given in Figures 1.52
and 1.53. Here again there are no empty argument positions in the
linked core.
‘Raising to subject’ constructions are made possible by M-atransitive verbs
like seem and appear. The LS for (143a) would be seem′ (Sally, [like′ (Sam,
wine)]) [MR Ø], in which neither of the semantic arguments of seem can be
Figure 1.52 Core cosubordination in (143a)
Figure 1.53 Core coordination in (143b)

realized as a direct core argument. From this LS two different patterns can
come, depending on whether there is a tense operator in the embedded LS. If
there is one, the result is It seems to Sally that Sam likes wine. If there is only one
tense operator over the entire LS, the result is (143a), in which the PSA of the
linked core appears as the PSA of the licensing core. This is a function of the
principles in (147b2) and (147c1) together with the M-atransitive nature of
seem: (147b2) requires that the linked core is missing a core argument
position, and (147c1) requires that the licensing core have at lease one core
argument position, which cannot be filled by one of the arguments of seem.
In order to avoid a Completeness Constraint violation, the ‘orphaned’ argu-
ment of like is linked to the open core argument slot in the licensing core,
yielding (143a). This is a different kind of argument sharing from that in
control constructions. It is a very common construction in the languages of
the world.
The ‘raising to object’ construction, on the other hand, is not. The LS for
(143b) would be believe′ (Sally, [like′ (Sam, wine)]), and as before, there are
two possible realizations, depending on whether there are two tense oper-
ators or just one over the entire LS. The first possibility yields Sally believes
that Sam likes wine, and the second yields (143b), which violates the
principle in (147b1). Believe and other ‘raising to object’ verbs are excep-
tional in not reducing their S-transitivity as required by (147b1), thereby
leaving an open core argument slot which can be filled by the ‘orphaned’
argument of the linked core verb, in this case like, thereby avoiding a
Completeness Constraint violation. The nature of argument sharing in this
construction is the same as that in the seem construction. (See also Van
Valin 2005, §7.3.2.)
The purpose construction in (142d), Sam brought wine for everyone to drink,
has an obligatory shared argument, unlike the rationale construction in
(154), which is a clausal juncture.
(154) a. Sam brought wine in order for Sally to drink it.

b. *Sam brought wine in order for Sally to drink.
c. Sam brought wine in order to drink it.
This construction does not allow a shared ‘object’ argument in the linked
clause, but does permit an optional omitted PSA coreferential with the PSA
in the first clause. The structure of (142d) is given in Figure 1.54; it is core
coordination, as shown by the fact that each core can take an independent
temporal positional modifier, for example Sam brought wine in the afternoon
for everyone to drink at the party during the evening.
The LS for this sentence is [do′ (Sam, [bring′ (Sam, winei)])] PURP [do′
(everyone, [drink′ (everyone, xi)])].68 This construction is significant, because
it involves a controller–pivot relationship in which neither the controller
nor the pivot is a traditional subject (see §1.3.2, Van Valin 2009). Because the
dependent LS is not an argument of the matrix LS, the reduction in
S-transitivity in the licensing core mandated by (147b1) does not apply.

Figure 1.54 Core coordination in (142d)
Despite the differences from the constructions in (142a–c), this construction

has an obligatory shared argument of the same type.
Thus, in nuclear junctures the arguments of the constituent nuclei are
pooled, just as in an LS of a multi-argument lexical verb. In non-subordinate
core junctures, there are two types of argument sharing. In control construc-
tions, one argument is shared between (or among) the cores; it occurs
overtly on one core and is co-indexed with its corresponding argument
position in the LSs of other cores. In ‘raising’ constructions, on the other
hand, a semantic argument from the linked core appears as a syntactic core
argument in the core headed by the ‘raising’ verb. In neither case is there an
empty syntactic slot in the linked core for the shared argument, nor is there
a null pronominal (e.g. PRO) representing the shared argument.
Argument sharing is only possible across core boundaries; across clause
boundaries argument identity involves coreference. Controller–pivot rela-
tions may involve argument sharing, as in obligatory control constructions,
or coreference, as in conjunction reduction constructions (e.g. (64), (65),
(113)). The following examples illustrate an important contrast.
(155) a. Sallyi was talking to Samj, and shei/hej waved to Margaret as she walked by.
b. Sallyi was talking to Samj, and proi/*j waved to Margaret as she walked by.
c. Sallyi was talking to Sam, and waved to Margaret as she walked by.
The first example involves coreference between overt pronouns, and one of
the major constraints in English is the requirement that third-person
pronouns must agree in gender in order to be interpreted as coreferential.
The second example lacks an overt pronoun in the PSA position in the

second clause, and most theories would posit null pronominal, pro, as the
PSA of the second clause. In (155b) only Sally can be interpreted as the one
who waved to Margaret; Sam is ruled out by a syntactic constraint on
coreference to the effect that in conjunction reduction constructions, the
PSA of the first clause must be the controller of the pivot (missing PSA) in
the second clause. It makes sense to talk of coreference between the
controller, RP Sally, and the phonologically null pro as pivot. But RRG does
not permit phonologically null lexical items in its syntactic representa-
tions, and accordingly (155b) is not a possible RRG analysis.69 Rather, the
structure RRG posits is (155c), with no null pronominal. This immediately
raises the question, what is Sally coreferential with? If there’s nothing in
the syntax instantiating the PSA of the second clause, how can this be a
case of coreference? Nevertheless, the PSA of the second clause is inter-
preted to be the same as the PSA of the first clause. The solution cannot be
formulated over the constituent projection of the LSC alone but must
include discourse. In Chapters 11 (Bentley) and 12 (Latrouite and Van Valin)
the RRG approach to the representation of discourse will be presented, and
it is essential to the analysis of conjunction reduction in particular and for
controller–pivot relations in clausal junctures in general. See Section 12.4
in Chapter 12 for detailed presentation of the RRG analysis of conjunction
reduction.
Controller–pivot relations in clausal junctures need not always be gram-
matically constrained, as they are in English, Icelandic, German, and Cro-
atian, for example. Archi, a Caucasian language (Kibrik 1979), lacks syntactic
or semantic constraints on controller–pivot coreference in clausal junctures.
The following examples from Kibrik (1979) illustrate this. In the structural
schemas, the italicized expressions with strikethrough do not occur overtly
in the sentence.
(156) a. Adamli [k’arāši xit’-boli] č’ele goıroı-abti.

man.erg down push-ger stone.abs roll-aor.
‘The man rolled the stone, having pushed it down.’
a′. man.erg [man.erg stone.abs push down] stone.abs roll
b. Os Helekulin lo [jab liqi’ili
one hen.gen child[IVsg].abs this[IIIsg].erg eagle[IIIsg].erg
oẍali] oqıali].
IVsg.carry_away.ger [IVsg].disappear
‘One of the chicks disappeared, carried away by an eagle.’
b′ chick.abs [eagle.erg chick.abs carry away] disappear
c. Bošor [k’oaHan soli] weır˙šu.
man[Isg].abs [stick[IVsg].abs hold.IVsg.ger] run.Isg.aor
‘The man ran, holding the stick.’
c′. man.abs [man.erg stick.abs hold] run
There are two controller–pivot relations in (156a): an ergative transitive

actor controlling an ergative transitive actor, and an absolutive transitive
undergoer controlling an absolutive transitive undergoer. In (156b) an

absolutive intransitive undergoer controls an absolutive transitive under-

goer, and in (156c) an absolutive intransitive actor controls an ergative
transitive undergoer. Thus intransitive and transitive actors can function
as controllers (a, c) and as pivots (a, c), and similarly intransitive and
transitive undergoers can function as controllers (a, b) and as pivots (a, b).
Thus there are no constraints on coreference other than that the controller–
pivot relations should make sense in the context (A. E. Kibrik, personal
communication).
In core and clausal junctures some of the constructions express argument
identity by only overtly instantiating the first of the possible occurrences of
the arguments. This is traditionally described in terms of ‘zero anaphora’.
Moreover, there may be structural constraints on the interpretation of
argument identity. It should be clear from this discussion that the nature
of ‘zero anaphora’ in core junctures is very different from that of ‘zero
anaphora’ in clausal junctures. This is summarized in (157).
(157) a. Argument-sharing [linking ‘zeros’] vs. coreference [discourse ‘zeros’]

b. Coreference: structurally constrained (155c) vs. structurally unconstrained
(156)
As shown in Figures 1.50 to 1.54, there is no gap or empty argument

position in the non-subordinate core junctures, and therefore describing
them in terms of ‘zero anaphora’ is quite incorrect from an RRG perspec-
tive. There are no zeros in these constructions. In the clausal junctures, on
the other hand, there are lacunae in the linked clauses; in (155c) there is no
PSA in the second clause, and in the Archi examples in (156) one or more
core arguments is missing from the linked clause. These lacunae are
analysed as zero elements in some approaches but not in RRG. What is
significant is the contrast between the structurally constrained corefer-
ence in English and many other languages and the structurally uncon-
strained coreference in Archi. With respect to argument sharing in core
junctures, the interpretation in (142a, b, c) is a function of the theory of
control, and in (142d) it follows from the constructional schema. The
argument sharing is constrained primarily lexically rather than by
morphosyntactic structure.
References
Aissen, Judith. 1987. Tzotzil Clause Structure. Dordrecht: Reidel.

Balogh, Kata, Anja Latrouite and Robert D. Van Valin, Jr. 2020. Nominal
anchoring: Introduction. In Kata Balogh et al. (eds.), Nominal Anchoring:
Specificity, Definiteness, and Article Systems Across Languages, 1–18. Berlin:
Language Sciences Press.
Beavers, John. 2010. The structure of lexical meaning: Why semantics really
matters. Language 86: 821–864.

Beavers, John. 2011. An aspectual analysis of ditransitive verbs of caused

possession in English. Journal of Semantics 28: 1–54.
Beavers, John. 2013. Aspectual classes and scales of change. Linguistics 51:
681–706.
Becker, Alton. 1975. A linguistic image of nature: The Burmese numerative
classifier system. Linguistics 13: 109–122.
Belloro, Valeria. 2007. Spanish Clitic Doubling: A Study of the Syntax–Pragmatics
Interface. PhD dissertation, University at Buffalo (SUNY) [available on RRG
website, https://rrg.caset.buffalo.edu/].
Bentley, Delia. 2006. Split Intransitivity in Italian. Berlin: Mouton.
Bentley, Delia. 2008. The interplay of focus structure and syntax: Evidence
from two sister languages. In Van Valin (ed.), 263–284.
quantized change. Paper presented at 2019 International Role and Reference
Grammar Conference, University at Buffalo (SUNY).
Bentley, Delia. 2022. Internally caused change as change by inner predispos-
ition: Comparative evidence from Romance. Ms., University of Manchester.
Berio, Leda, Anja Latrouite, Robert D. Van Valin, Jr. and Gottfried Vosgerau.
2017. Immediate and general common ground. In Patrick Brézillon, Roy
Turner and Carlo Penco (eds.), Modeling and Using Context, 1–14. Heidelberg:
Springer International Publishing.
Bickel, Balthasar. 2010. Capturing particulars and universals in clause link-
age: A multivariate analysis. In Bril (ed.), 51–101.
Boas, Franz and Ella Deloria. 1941. Dakota Grammar. Memoirs of the National
Academy of Sciences, Vol. 23. Washington, DC: US Govt. Printing Office.
Bohnemeyer, Jürgen. 2019. Operator, information: Revisiting the operator
projection in RRG, with special emphasis on tense, aspect, and finiteness.
Paper presented at the 2019 RRG Conference, University at Buffalo (SUNY).
Bohnemeyer, Jürgen and Robert D. Van Valin. Jr. 2017. The Macro-event Prop-
erty and the layered structure of the clause. Studies in Language 41: 142–197.
Bohnemeyer, Jürgen, Lindsay K. Butler and T. Florian Jaeger. 2016. Head-
marking and agreement: Evidence from Yucatec Maya. In Jens Fleisch-
hauer, Anja Latrouite and Rainer Osswald (eds.), Explorations of the Syntax–
Semantics Interface, 169–207. Düsseldorf: Düsseldorf University Press.
Bolinger, Dwight. 1975. Concept and percept: Two infinitive constructions
and their vicissitudes. In World Papers in Phonetics: Festschrift for Dr. Onishi
Kiju, 65–91. Tokyo: Phonetic Society of Japan.
Bresnan, Joan. 1982. Control and complementation. In Joan Bresnan (ed.), The
Mental Representation of Grammatical Relations, 282–390. Cambridge, MA: MIT
Press.
Bresnan, Joan and Sam A. Mchombo. 1987. Topic, pronoun and agreement in
ˇ
Chichewa. Language 63: 741–782.
Bril, Isabelle (ed.). 2010. Clause Linking and Clause Hierarchy. Amsterdam: John
Benjamins.

Cain, Bruce and James Gair. 2000. Dhivehi (Maldivian). Languages of the
World/Materials 63. Munich: Lincom Europa.
Chief, Liangchen. 2007. Scalarity and Incomplete Event Descriptions in Mandarin
Chinese. PhD dissertation, University at Buffalo (SUNY).
Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton.
Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT
Press.
Chomsky, Noam. 1970. Remarks on nominalization. In Roderick Jacobs and
Peter Rosenbaum (eds.), Readings in English Transformational Grammar,
184–221. Waltham, MA: Ginn & Co.
Chomsky, Noam. 1986. Knowledge of Language: Its Nature, Origin, and Use. New
York: Praeger.
Diedrichsen, Elke. 2004. The German ‘bekommen-passive’ and RRG. In Brian
Nolan (ed.), Linguistic Theory and Practice: Description, Implementation and
Processing. Proceedings of the 2004 RRG Dublin Conference, 49–72.
Diedrichsen, Elke. 2011. The theoretical importance of constructional sche-
mas in RRG. In Nakamura (ed.), 169–197.
Dixon, R. M. W. 1972. The Dyirbal Language of North Queensland. Cambridge:
Dixon, R. M. W. (ed.). 1976. Grammatical Categories in Australian Languages.
Canberra: Australian Institute of Aboriginal Studies.
Dixon, R. M. W. 1984. The semantic basis of syntactic properties. BLS 10:
583–595.
Dixon, R. M. W. 1991. A New Approach to English Grammar, on Semantic Principles.
Oxford: Oxford University Press.
Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel.
Dryer, Matthew. 1986. Primary objects, secondary objects, and antidative.
Language 62: 808–845.
Everett, Caleb. 2006. Patterns in Karitiâna: Articulation, Perception, and Grammar.
PhD dissertation, Rice University, TX [available on Information Structure in
Amazonian languages website].
Everett, Caleb. 2008. Constituent Focus in Karitiâna. Unpublished ms. [available
on Information Structure in Amazonian languages website].
Everett, Daniel. 2002. Notes on an RRG Theory of Morphology. Unpublished ms.
Everett, Daniel. 2005. Cultural constraints on grammar and cognition in
Piraha:~ Another look at the design features of human language. Current
Anthropology 76: 621–646.

Everett, Daniel. 2008. Wari’ intensional state constructions. In Van Valin
(ed.), 381–412.
Everett, Daniel and Doris Kern. 1997. Wari’. London: Routledge.
Fleischhauer, Jens. 2016. Degree Gradation of Verbs. Doctoral dissertation,
Heinrich Heine University Düsseldorf.
Foley, William A. 2010. Clause linkage and nexus in Papuan languages. In
Bril (ed.), 27–50.

Franklin, Karl. 1971. A Grammar of Kewa, New Guinea. (Pacific Linguistics C-16).
Canberra: Australian National University.
Goldberg, Adele. 2009. Constructions at work. Cognitive Linguistics 20:
201–224.
González Vergara, Carlos. 2006. Las construcciones no reflexivas con ‘se’: Una
propuesta desde la Gramática del Papel y la Referencia. PhD dissertation, Uni-
versidad Complutense de Madrid [available on the RRG website, https://
rrg.caset.buffalo.edu/].
González Vergara, Carlos. 2009. One rule to rule them all: Logical structures
for Spanish non-reflexive se sentences. In Guerrero Valenzuela, Ibáñez and
Belloro (eds.), 361–379.
González Vergara, Carlos. 2011. Se-incompatible predicates in Spanish:
A RRG explanation. In Nakamura (ed.), 134–142.
Guerrero Valenzuela, Lilián, Sergio Ibáñez and Valeria Belloro (eds.). 2009.
Studies in Role and Reference Grammar. México: UNAM.
Guerrero Valenzuela, Lilián and Robert D. Van Valin, Jr. 2004. Yaqui and the
analysis of primary object languages. International Journal of American Lin-
guistics 70: 290–319.
Hale, Kenneth. 1976. The adjoined relative clause in Australia. In Dixon (ed.),
78–105.
Hasegawa, Yoko. 1996. A Study of Japanese Clause Linkage: The Connective TE in
Japanese (Studies in Japanese Linguistics 5). Stanford, CA: CSLI Publications.
Haviland, John. 1979. How to talk to your brother-in-law in Guugu Yimid-
hirr. In T. Shopen (ed.), Languages and Their Speakers, 160–239. Cambridge,
MA: Winthrop.
Himmelmann, Nikolaus. 2008. Lexical categories and voice in Tagalog. In
Paul Kroeger (ed.), Voice and Grammatical Relations in Austronesian Languages,
247–293. Stanford, CA: CSLI Publications.
Holisky, Dee Ann. 1987. The case of the intransitive subject in Tsova-Tush
(Batsbi). Lingua 71: 103–132.
Hudson, Richard. 1992. So-called ‘double objects’ and grammatical relations.
Language 68: 251–276.
Humboldt, Wilhelm von. 1836. Über die Verschiedenheit des menschlichen Sprach-
baues und ihren Einfluss aft die geistige Entwicklung des Menschengeschlects.
Berlin: Dümmler.
Hyman, Larry and Karl Zimmer. 1976. Embedded topic in French. In Charles
Li (ed.), 189–211.
Ikegami, Yoshihiko. 1985. ‘Activity’ – ‘accomplishment’ – ‘achievement’ –
A language that can’t say ‘I burned it, but it didn’t burn’ and one that can.
In A. Makkai and A. Melby (eds.), Linguistics and Philosophy: Essays in honor of
Rulon S. Wells, 265–304. Amsterdam: John Benjamins.
Imai, Shingo. 1998. Logical Structures and Case Marking in Japanese. MA thesis,
University at Buffalo (SUNY) [available on RRG website, https://rrg.caset.
buffalo.edu/].

Jackendoff, Ray. 1990. Semantic Structures. Cambridge, MA: MIT Press.

Jackendoff, Ray. 2002. Foundations of Language: Brain, Meaning, Grammar, Evo-
lution. Oxford: Oxford University Press.
Jelinek, Eloise. 1984. Empty categories and non-configurational languages.
Natural Language and Linguistic Theory 2: 39–76.
Johnson, Mark. 1987. A New Approach to Clause Structure in Role and Reference
Grammar. Davis Working Papers in Linguistics 1, 55–59. Davis: University
of California.
Kailuweit, Rolf. 2008. ‘Floating plurals’, pro drop and agreement: An
optimality-based RRG approach. In Van Valin (ed.), 179–202.
Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press.
Kibrik, Alexander E. 1979. Canonical ergativity and Daghestan languages. In
Frans Planck (ed.), Ergativity, 61–78. London: Academic Press.
Kihara, C. Patrick. 2017. Aspects of G~ıkuy
~ u~ (Kikuyu) Complex Sentences: A Role and
Reference Grammar Analysis. PhD dissertation, Heinrich Heine University

Düsseldorf [available on RRG website, https://rrg.caset.buffalo.edu/].
Kolmer, Agnes. 1998. Pluralität im Tagalog (Arbeitspapiere des Instituts für
Sprachwissenschaft der Universität zu Köln, 31.) Cologne: Institut für
Sprachwissenschaft, Universität zu Köln.
Kuno, Susumu. 1973. The Structure of the Japanese Language. Cambridge, MA:
MIT Press.
Lambrecht, Knud. 1994. Information Structure and Sentence Form. Cambridge:
Langacker, Ronald. 2009. Cognitive (Construction) Grammar. Cognitive Lin-
guistics 20: 167–176.
Latrouite, Anja and Robert D. Van Valin, Jr. 2014. Event existentials in
Tagalog: A Role and Reference Grammar account. In I. Wayan Arka and
N. L. K Mas Indrawati (eds.), Argument Realizations and Related Constructions in
Austronesian Languages, Papers from 12-ICAL, Vol. 2, 161–174. Canberra: Asia-
Pacific Linguistics [available on RRG website, https://rrg.caset.buffalo.edu/].
Latrouite, Anja and Robert D. Van Valin, Jr. 2021. An RRG account of aspects
of the information structure-syntax interface in Tagalog. In Van Valin
(ed.), 257–283.
Li, Charles (ed.). 1976. Subject and Topic. New York: Academic Press.
Mann, William C. and Sandra A. Thompson. 1988. Discourse Description.
Matasović, Ranko. 2008. Transitivity in Kabardian. In Van Valin (ed.), 59–73.
Matasović, Ranko. 2009. A new typology of control constructions within the
framework of Role and Reference Grammar. In Guerrero Valenzuela,
Ibáñez and Belloro (eds.), 305–318.
Matasović, Ranko. 2010. Kabardian causatives, reflexives and case marking
domains. Suvremena Lingvistika 69: 45–63.
Matasović, Ranko. 2017. Agreement in Role and Reference Grammar:
A typology of possible targets. Suvremena Lingvistika 84: 157–171.
Nakamura, Wataru (ed.). 2011. New Perspectives in Role and Reference Grammar.
Newcastle upon Tyne: Cambridge Scholars Publishing.

Nakamura, Wataru. 2021. A neo-Jakobsonian account of default cases:

Instrumental vs. dative. In Van Valin (ed.), 135–168.
Narasimhan, Bhuvana. 1998. A lexical semantic explanation for ‘quirky’ case
marking in Hindi. Studia Linguistica 52: 48–76.
Nichols, Johanna. 1986. Head-marking and dependent-marking grammar.
Language 62: 56–119.
Nichols, Johanna, David A. Peterson and Jonathan Barnes. 2004. Transitivizing
and detransitivizing languages. Linguistic Typology 8: 149–211.
Nolan, Brian and Elke Diedrichsen (eds.). 2013. Linking Constructions into
Functional Linguistics: The Role of Constructions in Grammar. Amsterdam: John
Benjamins.
Nuhn, Patrick. 2021. Ay-Inversion in Tagalog: Information Structure and Morpho-
syntax of an Austronesian Language. Düsseldorf: Düsseldorf University Press
[available on RRG website, https://rrg.caset.buffalo.edu/].
Nunes, Mary L. 1993. Argument linking in English derived nominals. In Van
Valin (ed.), 375–432.
O’Connor Rob. 2008. A prosodic projection for Role and Reference Grammar.
In Van Valin (ed.), 227–244.
Ohori. Toshio. 2001. Some Thoughts on a New Systematization of Interclausal
Semantic Relations. Unpublished ms. [available on RRG website, https://rrg.
caset.buffalo.edu/].
Osborne, C. R. 1974. The Tiwi Language. Canberra: Australian Institute of
Aboriginal Studies.
Osswald, Rainer. 2021. Activities, accomplishments and causation. In Van
Valin (ed.), 3–30.
Park, Ki-seong. 1993. Korean Causatives in Role and Reference Grammar. MA
thesis, University at Buffalo (SUNY) [available on RRG website, https://rrg.
caset.buffalo.edu/].
Pensalfini, Rob. 2004. Towards a typology of non-configurationality. Natural
Language and Linguistic Theory 22: 359–408.
Pinker, Steven. 1989. Learnability and Cognition. Cambridge, MA: MIT Press.
Polinsky, Maria and Eric Potsdam. 2002. Backward control. Linguistic Inquiry
33: 245–282.
Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, MA: MIT Press.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik. 1972.
A Grammar of Contemporary English. New York: Seminar Press.
Rappaport Hovav, Malka and Beth Levin. 2008. The English dative
alternation: The case for verb sensitivity. Journal of Linguistics 44: 129–167.
Rijkhoff, Jan. 1990. Explaining word order in the noun phrase. Linguistics 28:
5–42.
Rijkhoff, Jan. 2002. The Noun Phrase: A Typological Study of Its Form and Structure.
Rögnvaldsson, Eiríkur. 1982. We need (some kind of a) rule of conjunction
reduction. Linguistic Inquiry 13: 557–561.
Ross, John Robert. 1967. Constraints on Variables in Syntax. PhD dissertation,
MIT.

Ruhnau, Arne. 2011. Interpretation of the Topological Field Model of the German Clause
in Terms of Role and Reference Grammar. MA thesis, Heinrich Heine University
Sag, Ivan A. 2012. Sign-Based Construction Grammar: An informal synopsis.
In Hans C. Boas and Ivan A. Sag (eds.), Sign-Based Construction Grammar,
Schachter, Paul. 1976. The subject in Philippine languages: Actor, topic,
actor-topic, or none of the above. In Li (ed.), 491–518.
Schachter, Paul and Fe Otanes. 1972. Tagalog Reference Grammar. Berkeley:
University of California Press.
Schwartz, Linda. 1993. On the syntactic and semantic alignment of attribu-
tive and identificational constructions. In Van Valin (ed.), 433–463.
Shimojo, Mitsuaki. 1995. Focus Structure and Morphosyntax in Japanese: wa and
ga, and Word Order Flexibility. PhD dissertation, University at Buffalo (SUNY)
[available on the RRG website, https://rrg.caset.buffalo.edu/].
Shimojo, Mitsuaki. 2011. The left periphery and focus structure in Japan-
ese. In Wataru Nakamura (ed.), New Perspectives in Role and Reference Gram-
mar, 266–293. Newcastle upon Tyne: Cambridge Scholars Publishing.
Silverstein, Michael. 1976. Hierarchy of features and ergativity. In Dixon
(ed.), 112–171.
Smith, Carlotta. 1997. The Parameter of Aspect (2nd ed.). Dordrecht: Reidel.
Talmy, Leonard. 2000. Toward a Cognitive Semantics. Cambridge, MA: MIT Press.
Toratani, Kiyoko. 2002. The Morphosyntax and Logical Structures of Compound
Verbs in Japanese. PhD dissertation, University at Buffalo (SUNY) [available
on RRG website, https://rrg.caset.buffalo.edu/].
Ullrich, Jan. 2011. New Lakota Dictionary (2nd ed.). Bloomington, IN: Lakota
Language Consortium.
Ullrich, Jan. 2018. Modification, Secondary Predication and Multi-Verb Construc-
tions in Lakota. PhD dissertation, Heinrich Heine University Düsseldorf
[available on RRG website, https://rrg.caset.buffalo.edu/].
Ullrich, Jan and Ben Black Bear, Jr. 2018. Lakota Grammar Handbook (2nd ed.).
Bloomington, IN: Lakota Language Consortium.
Van Hooste, Koen. 2018. Instruments and Related Concepts at the Syntax–Semantics
Interface. PhD dissertation, Heinrich Heine University Düsseldorf [available
on RRG website, https://rrg.caset.buffalo.edu/].
Van Valin, Robert D., Jr. 1977. Aspects of Lakhota Syntax. PhD dissertation,
University of California, Berkeley.
Van Valin, Robert D., Jr. 1985. Case marking and the structure of the Lakhota
clause. In Johanna Nichols and Anthony Woodbury (eds.), Grammar Inside
and Outside the Clause, 363–413. Cambridge: Cambridge University Press.
Van Valin, Robert D., Jr. 1990a. Semantic parameters of split intransitivity.
Language 66: 221–260.
Van Valin, Robert D., Jr. 1990b. Functionalism, anaphora and syntax. (Review
article on S. Kuno, Functional Syntax.) Studies in Language 14: 169–219.

Van Valin, Robert D., Jr. 1991. Another look at Icelandic case marking and
grammatical relations. Natural Language and Linguistic Theory 9: 145–194.
Van Valin, Robert D., Jr. (ed.). 1993. Advances in Role and Reference Grammar.
Van Valin, Robert D., Jr. 1993. A synopsis of Role and Reference Grammar. In
Van Valin (ed.), 1–164.
Van Valin, Robert D., Jr. 1998. The acquisition of wh-questions and the
mechanisms of language acquisition. In Michael Tomasello (ed.), The New
Psychology of Language: Cognitive and Functional Approaches to Language Struc-
ture, 221–49. Mahwah, NJ: LEA [available on RRG website, https://rrg.caset.
buffalo.edu/].
Van Valin, Robert D., Jr. 1999. Generalized semantic roles and the syntax–
semantics interface. In F. Corblin, C. Dobrovie-Sorin and J.-M. Marandin,
(eds.), Empirical Issues in Formal Syntax and Semantics 2, 373–389. The Hague:
Thesus [available on RRG website, https://rrg.caset.buffalo.edu/].
Van Valin, Robert D., Jr. 2001. The acquisition of complex sentences: A case
study in the role of theory in the study of language development. Chicago
Linguistic Society Proceedings 36(2): 511–531.
Van Valin, Robert D., Jr. 2002. The development of subject–auxiliary inver-
sion in English wh-questions: An alternative analysis. Journal of Child Lan-
guage 29: 161–175.
Van Valin, Robert D., Jr. 2003. Some remarks on the bekommen-‘passive’ in
German. Invited paper presented at the 9th German Linguistics Association
Conference, University at Buffalo (SUNY).
Van Valin, Robert D., Jr. 2004. Semantic macroroles in Role and Reference
Grammar. In Rolf Kailuweit and Martin Hummel (eds.), Semantische Rollen,
62–82. Tübingen: Narr [available on RRG website, https://rrg.caset.buffalo.
edu/].
Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D., Jr. 2006. Semantic macroroles and language processing.
In Ina Bornkessel, Matthias Schlesewsky, Bernard Comrie and Angela D.
Friederici (eds.), Semantic Role Universals and Argument Linking: Theoretical,
Typological, and Psycholinguistic Perspectives, 263–301. Berlin: Mouton de
Gruyter.
Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics–
Pragmatics Interface. Amsterdam: John Benjamins.
categories in RRG. In Van Valin (ed.), 161–178.
Van Valin, Robert D., Jr. 2009. Privileged syntactic arguments, pivots, and
controllers. In Guerrero Valenzuela, Ibáñez and Belloro (eds.), 45–68.
Van Valin, Robert D., Jr. 2012a. Lexical representation, co-composition, and
linking syntax and semantics. In James Pustejovsky, Pierrette Bouillon,
Hitoshi Isahara, Kyoko Kanzaki and Chungmin Lee (eds.), Advances in
Generative Lexicon Theory, 67–107. Dordrecht: Springer.

Van Valin, Robert D., Jr. 2012b. Some issues in the linking between syntax
and semantics in relative clauses. In Bernard Comrie and Zarina Estrada
Fernández (eds.), A Typological Overview of Relative Clauses in Languages of the
Americas, 47–64. Amsterdam: John Benjamins.
Timberlake (eds.), Language Typology and Historical Contingency. In honor of
Johanna Nichols, 91–124. Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 2016. An overview of information structure in three
Amazonian languages. In M. M. Jocelyne Fernandez-Vest and Robert D. Van
Valin, Jr. (eds.), Information Structuring of Spoken Language from a Cross-
Linguistic Perspective, 77–92. Berlin: Mouton de Gruyter.
Van Valin, Robert D., Jr. 2018a. Some issues regarding (active) accomplish-
ments. In Rolf Kailuweit, Lisa Künkel and Eva Staudinger (eds.), Applying
and Expanding Role and Reference Grammar, 71–94. Freiburg: FRIAS.
Van Valin, Robert D., Jr. 2018b. Dative case and oblique subjects. In Jóhanna
Barðdal, Steven Carey, Thórhallur Eythórson and Na’ama Pat-El (eds.), Non-
Canonically Case-Marked Subjects: The Reykjavík-Eyjafjallajökull Papers, 115–131.
Van Valin, Robert D., Jr. (ed.). 2021. Challenges at the Syntax–Semantics–
Pragmatics Interface: A Role and Reference Grammar Perspective. Newcastle upon
Tyne: Cambridge Scholars Publishing.
Van Valin, Robert D., Jr. 2021. Cosubordination. In Van Valin (ed.), 241–283.
Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning
Van Valin, Robert D., Jr. and David P. Wilkins. 1993. Predicting syntactic
structure from semantic representations: remember in English and
Mparntwe Arrernte. In Van Valin (ed.), 499–534.
Van Valin, Robert D., Jr. and David P. Wilkins. 1996. The case for ‘effector’:
Case roles, agents and agency revisited. In M. Shibatani and S. Thompson
(eds.), Grammatical Constructions, 289–322. Oxford: Oxford University Press.
Press.
Walton, Charles. 1986. Sama Verbal Semantics: Classification, Derivation and
Inflection. Manila: Linguistic Society of the Philippines.
Watters, James K. 1993. An investigation of Turkish clause linkage. In Van
Valin, (ed.), 535–560.
Winther-Nielsen, Nicolai. 1995. A Functional Discourse Grammar of Joshua:
A Computer-Assisted Rhetorical Structure Analysis. Stockholm: Almqvist &
Wiksell.
Winther-Nielsen, Nicolai. 2021. Why Eve shouldn’t eat the snake: An intelli-
gent answer from corpus-driven information structure and reference
tracking in Biblical Hebrew. In Van Valin (ed.), 285–307.
Yang, Byong-seon. 1994. Morphosyntactic Phenomena of Korean in Role and Refer-
ence Grammar: Psych-Verb Constructions, Inflectional Verb Morphemes, Complex

Sentences, and Relative Clauses. PhD dissertation, University at Buffalo

(SUNY). Published by Hankuk Publishers, Seoul, 1994.
Zaenen, Annie, Joan Maling and Hölskuldur Thráinsson. 1985. Case and
grammatical functions: The Icelandic passive. Natural Language and Linguis-
tic Theory 3: 441–483.
Notes
1 In the tree diagrams modifiers are related to the units modified via
arrows. Kallmeyer and Osswald (Chapter 20) propose an alternative for-
malization of the layered structure of the clause which does not use
arrows to represent adjunct modifiers. I will continue to use the trad-
itional RRG representations.
2 Bohnemeyer and Van Valin (2017) show that the core is the unit of the LSC
which expresses a single event; it has the macro-event property, which the
nucleus and the clause lack. See Section 1.3.4.4.
3 For explications of these notions, see Bentley (Chapter 11).
4 The auxiliary did is not attached to anything in the tree in Figure 1.4. Its
status will be addressed in Section 1.2.3. With respect to the PrCS and
resumptive elements, the claim that PrCS XPs never co-occur with a
resumptive element holds true for English and many other languages.
However, in head-marking languages (see Section 1.2.7.2) which have a
PrCS, e.g. Tzotzil (Aissen 1987), an argument in the PrCS is cross-
referenced on the verb, which could be viewed as a type of resumptive
element; what would be impossible would be for there to be an inde-
pendent pronoun in the clause referring to the PrCS RP. Spanish pre-
sents an interesting case as well. Dative arguments are obligatorily
realized by a dative clitic on the verb and optionally also by a dative-
marked noun or pronoun (see (58) in Section 1.2.7.2); in wh-questions in
which the wh-expression is dative, such as (58d), it occurs in the PrCS
and is still cross-referenced by the dative clitic in the core. Hence the
issue of resumption and the PrCS is more complex than it appears to be
in English.
5 It’s been claimed by a number of linguists, most notably Kayne (1994) in
his theory of the antisymmetry of syntax, that all displacement is to the
beginning of the sentence. These Dhivehi examples are a counterexample
to this claim.
6 I am grateful to colleagues from Iran (Dr Mozhgan Neisani, Dr Zahra
Ghane’) for bringing this problem to my attention. See Winther-Nielsen
(2021) for the same problem regarding the analysis of Biblical Hebrew. The
post-detached position (PoDP) involves an intonation break and resump-
tive cataphoric pronoun for arguments, e.g. What did shei eat after lunch,
Maryi?

7 ‘DetP’ will be used for ‘determiner phrase’ rather than the usual ‘DP’,
because ‘DP’ has already been used to stand for ‘detached position’ in
PrDP and PoDP.
8 See Section 1.2.7.2 for an explanation as to why the independent RP is
not core-internal.
9 In Van Valin (2005) ‘RP’ stands for ‘referential phrase’, but this term is
problematic for various reasons (see Peterson, Chapter 2 of this volume).
In Van Valin (2008) it was replaced by ‘reference phrase’ and defined as a
potentially referring expression, thereby taking into account the fact that
RPs can be non-referential in certain contexts.
10 RP, MP and nucleus are universal categories, while PP is not, since there
are languages which lack adpositions altogether, e.g. Dyirbal.
11 See Peterson (this volume, Chapter 2) for an in-depth discussion of
this topic.
12 Peterson (Chapter 2) argues that Kharia, a South Munda language (India),
makes no categorial distinctions among content morphemes at all.
13 The existence of attributive modifiers in Lakhota has been controversial
since Boas and Deloria (1941). See Ullrich (2018) for a detailed analysis of
modification in Lakhota.
14 Bohnemeyer (2019) presents an alternative conception of operators and
their relationship to the LSC based on a different theory of tense and
aspect from the one assumed traditionally in RRG.
15 The formalization of RRG proposed in Kallmeyer and Osswald (Chap-
ter 20) does away with the operator projection and represents both
constituents and operators in the same tree structure. I will continue
to use the traditional representations.
16 Lakhota lacks tense marking, so this sentence can be interpreted as
present or past, depending on context.
17 This is not the complete array; see Ullrich (2011: 820–822) for a full list.
18 See Peterson (Chapter 2) for more discussion and exemplification of
this claim.
19 See O’Connor (2008) for the outlines of a prosodic projection for the LSC
in which the intonational properties of different sentence types and
constructions can be represented.
20 Additional examples of attachment to more than one projection will be
given later in this section and in Section 3.
21 See Toratani (Chapter 9) for an in-depth discussion of adverbs and Ibáñez
Cerda (Chapter 10) for a detailed discussion of adpositional phrases.
22 If the adjunct PP were treated as a parenthetical, then Max drank, after the
argument with Mary, an entire bottle of wine would be acceptable, but it is
highly marked and not the same structure as (30b).
23 For detailed discussion of the ordering of constraints on adverbs, see
Toratani (Chapter 9).
24 Unlike Japanese, German is not consistently left-branching; it has many
more prepositions than postpositions, and in complex sentences

dependent units follow the main clause (default), as in English and

unlike Japanese, where they precede it.
25 Ross (1967) proposed treating these phrases as NPs, rather than PPs, since
in a language like German, for example, a case-marked phrase headed by
a noun would be analysed as an NP, not a PP. This is problematic,
however, because it fails to distinguish direct core arguments from
oblique core arguments, a distinction relevant to important
grammatical phenomena.
26 The CSP node in the RP applies to languages like Tagalog, Japanese and
Korean which have case markers that are not bound morphemes but are
not adpositions, either. In Korean and Japanese, for example, the distinc-
tion between case marker and postposition is reflected in the ability of
case particles to be omitted and the inability of postpositions to
be dropped.
27 As noted in Van Valin (2013: 93), the recognition of the distinctiveness of
head-marking languages goes back at least as far as Wilhelm von
Humboldt (1836).
28 The -A signifies that the final vowel undergoes ablaut in certain environ-
ments. See Ullrich (2011: 754).
29 There are varieties of Latin American Spanish which allow (57c); see
Belloro (2007).
30 Foley (2010) and Bickel (2010) reject the notion of cosubordination as a
valid linkage type. Van Valin (2021) shows that their arguments are based
on an outmoded version of the LSC and make assumptions about
cosubordination which are incorrect. In fact, a reanalysis of their data
supports the validity of cosubordination as a nexus relation.
31 The term ‘subject’ is in single quotes because it has no theoretical status
in RRG. This construction, usually called conjunction reduction,
contains a controller–pivot (the missing RP) relationship; see Sections
1.7.2 and 12.4 for further discussion of conjunction reduction. See
LaPolla (Chapter 5) for a presentation of the RRG approach to
grammatical relations.
32 See París (Chapter 15) for an analysis of it-clefts.
33 These discussions assume the previous version of the IRH presented in
Van Valin (2005), but the issues relating to the many-to-one mapping
between the two sides of the IRH are not affected by the revisions
proposed here.
34 Let occurs in the passive voice in some limited combinations, such as let
go meaning ‘fired’, e.g. He was let go after showing up at work drunk. Note
there are no comparable combinations with come, e.g. *He was let come to
the party.
35 See Van Valin (1990b), Van Valin and LaPolla (1997, §7.5), and Van Valin
(2005, §§5.3, 7.5), for detailed discussions of reflexivization in English.
36 The representations are meant as ‘generic’ junctures, like those in (59),
because the issue of nexus will be addressed separately below.

37 ‘DO’ indicates that the argument is an agent with an activity predicate,

and ‘do′’ is an unspecified activity predicate used in accomplishment
representations which do not specify the nature of the causing activity,
e.g. in ‘The cat killed the rat’, the cat’s killing action is unspecified.
38 See Osswald (2021) for discussion of Dowty’s own revisions of the
decompositional system.
39 The term ‘accomplishment’ has been used in a number of different
senses by Vendler and Dowty and in RRG; see Van Valin (2018a) for
detailed discussion.
40 The contrast with respect to inherent endpoints is rendered in English by
simple past vs. past progressive, but it is not an inflectional opposition.
See Van Valin (2018a) for examples from other languages which do not
involve inflectional distinctions. Also, the addition of processes to the
system was proposed in Van Valin (2005).
41 Because stumble into the room is what Talmy (2000) has termed a ‘satellite-
framed’ expression and enter the room stumbling is a ‘verb-framed’ expres-
sion, they are composed differently. Stumble is a manner of motion verb,
and so its LS is do′ (x, [move.manner′ (x)]), and the remainder of the LS is
supplied by the PP into the room (the satellite). In the verb-framed enter the
room stumbling, the motion predicate is unspecified, and the verb contrib-
utes the path and goal, with the manner component added, as in (81a2ii).
In Van Valin (2018a) it was argued that an expression like stumble into the
room is a nuclear juncture, and this suggests that enter the room stumbling
could be one as well.
42 This also implies that the LS for activity verbs which potentially take an
incremental theme or path would have a different LS from those that do
not, e.g. shine (do′ (x, [shine′ (x)])), vs. run (do′ (x, [run′ (x)]) ^ PROC
covering.path.distance′ (x, (y))). Note that run’s LS would be different
from stumble’s, because Bill stumbled doesn’t necessarily involve him
covering any distance, but Bill ran does.
43 Two points need to be kept in mind. First, the elements of the decom-
position which are in boldfaceþprime, e.g. do′, or in all caps, e.g. CAUSE,
INGR, are part of a semantic metalanguage and are not English, despite
the obvious similarities. Second, when representing the LSs of sentences
from other languages, only the lexical items from the language which
function as arguments should appear in LSs. In a sentence like (13a) from
Lakhota, for example, the LS would be do′ (wíŋyaŋ kiŋ hená, [dance′
(wíŋyaŋ kiŋ hená)]) (plus operators) ‘Those women dance’, not ečhúŋ′
(wíŋyaŋ kiŋ hená, [wačhí′ (wíŋyaŋ kiŋ hená)]. See Van Valin (2005: 45–46),
esp. footnote 9.
44 Three things need to be noted with respect to causation. First, the
causatives under discussion do not include so-called ‘internal causatives’
associated with verbs like (intransitive) rot or bloom. Causation involves a
causer and a causee, and the single argument of these verbs is neither,
hence they cannot be analysed as causative in the same sense as the

examples in (91g). These verbs can be better described as involving ‘inner

force’, a notion expressed in e.g. Lakhota grammar by the instrumental
prefix na-, e.g. -pȟópA ‘to be popped or burst open’ þ na- ¼ napȟópA ‘to
burst with explosive force from within, burst or pop open by itself or
from inner force, to explode; erupt (as a volcano)’ (Ullrich 2011: 485, 386)
(see Bentley 2022 for verbs of inner predisposition). Second, representing
causation by CAUSE is a gross oversimplification, since it fails to take
into account the different types of causality in the IRH found in (81a, b,
e). A first step toward a more sophisticated approach to causation can be
seen in Van Hooste’s (2018) proposal for incorporating the theory of force
dynamics proposed in Talmy (2000) into RRG. Third, it is well known that
in some languages (e.g. Mandarin, Chief 2007; Japanese, Ikegami 1985;
Korean, Park 1993) causatives don’t necessarily entail that the causation
has a result state, which yields apparently contradictory sentences trans-
lated as ‘I killed him but he didn’t die’, ‘I broke it but it didn’t break’.
Chief (2007) shows that the predicates in such examples are scalar in
nature, unlike their English counterparts, which means they involve
non-quantized changes of state of the melt variety (not the widen variety).
In order to express a quantized change of state with a specific result
state, the result state must be explicitly realized in some way, in which
case there is a result state, and the resulting sentences are contradictory
like their English translations. See Van Valin (2005: 42), fn. 5.
45 For detailed discussion, see Van Valin and LaPolla (1997), §4.7.3, Van
Valin (2005), §2.3.
46 The difference between (94a2ii), e.g. Bill entered the room stumbling, and
(94b4iii), e.g. Bill sat reading the newspaper, lies in the fact that the stum-
bling is an integral part of the motion, as can be seen from the fact that
it’s possible to integrate stumble into the motion predication, i.e. Bill
stumbled into the room. Accordingly, the motion cannot be separated out
as a separate action; in other words, no stumbling, no motion. Sitting is
not part of the action of reading, on the other hand, as one can read
sitting, standing or lying, hence the stance verb represents an independ-
ent action from the reading.
47 Russian has a lexical verb of possession, imet’, which is subject to restric-
tions on its use and is not a colloquial expression of possession. This
contrasts with Croatian imati ‘have’, which would be used in sentences
like (95). (R. Matasović, personal communication).
48 The Russian verb imeti ‘have, possess’ can be used in existential construc-
tions. (R. Matasović, personal communication).
49 Semantic macroroles are an original RRG construct; see Van Valin (1999)
for a historical overview and a comparison with analogous ideas in
other approaches.
50 All of the x-arguments with activity predicates are subtypes of the basic
notion of effector, the doer of the action, which may or may not be
animate or volitional.

51 An exception is found in the present-tense series of Georgian verbs,

which have a nominative [A]–dative [U] pattern with transitive verbs
(Van Valin 1990a).
52 In some languages, e.g. Lakhota, Yaqui (Uto-Aztecan), Belhare (Sino-
Tibetan), the primary pattern with three-argument verbs is (103a) with
the recipient as the undergoer, rather than (103a′) as in German, Cro-
atian and many other languages (see Dryer 1986). Guerrero and Van
Valin (2004) proposed a revision to the AUH to account for this: a verb
may follow Principle A (lowest-ranking argument), Principle B (second
highest ranking), or Principle C (both A and B are possible). Most three-
place verbs in English follow C, but there are some that are restricted to
A, e.g. put, donate. A possible example of an English verb which follows
only B is envy. Yaqui has all three possibilities, with B the predominant
pattern (see Section 1.6.4.2). German signals the different options via
derivational morphology, e.g. schenken ‘give as a gift’(¼ (100c)), laden
‘load’(¼(100b)) (A) vs. beschenken (¼(100c’)), beladen (¼(100b’))(B).
53 The reason seem′ (x, y) is [MR 0] is that neither of the arguments can
appear as a direct core argument in the core headed by seem. The
x-argument can only be realized in a to-PP, and the y-argument can
only appear as an extraposed that-clause, e.g. seem′ (Bill, [like′ (Mary,
pizza)]) —> It seems to Bill that Mary likes pizza, *Bill seems that Mary likes
pizza. Note that in Mary seems to like pizza, Mary is not a semantic
argument of seem but rather is one of like. This construction will be
discussed in Section1.7.
54 O’Connor’s (2008) prosodic projection makes possible the explicit repre-
sentation of the prosodic properties of utterances. This is very relevant to
information structure, but it is a separate projection from the infor-
mation structure projection, because there are languages which appear
not to use prosody in marking information structural contrasts, e.g.
Karitiâna (see C. Everett 2006, 2008; Van Valin 2016). In addition, prosody
may interact directly with the operator projection, e.g. marking IF con-
trasts, and with the constituent projection, e.g. marking grouping
of constituents.
55 The ‘proi’ is for expository purposes only and would not be part of an
RRG analysis of conjunction reduction. See Section 1.7.2; Chapter 12,
Section 12.4.
56 The qualification ‘highest-ranking macrorole core argument’ is necessary,
because in a passive construction, the actor, which is the highest-ranking
macrorole, appears as an adjunct in the core-level periphery, which
leaves the undergoer as the highest-ranking macrorole core argument
and therefore the controller of person agreement and the controller in
conjunction reduction. See Matasović (2017) for a presentation of an
RRG-based theory of agreement.
57 RRG is thus a type of construction grammar, albeit one very different
from ‘mainstream’ Construction Grammar, e.g. Sign-Based Construction

Grammar (Sag 2012), which is very close to HPSG, or Cognitive Construc-

tion Grammar (Goldberg 2009; Langacker 2009). What unites the main-
stream approaches is the assumption that constructions are stored
pairings of form and meaning and therefore that everything from a
bound affix, to a word, to a phrase, to a passive are constructions, and
consequently the lexicon–grammar distinction is a false dichotomy,
with lexical items just being simpler form–meaning pairings than gram-
matical constructions. One inevitable consequence of this assumption is
that all constructions are language-specific, and accordingly it is
extremely difficult, if not impossible, to capture any cross-linguistic
generalizations. RRG does not make this assumption: constructions in
RRG are not stored form–meaning pairings but rather compilations of
language-specific applications of the general principles of the theory.
Numerous examples will be given in the text. Hence cross-linguistic
generalizations are readily captured in the RRG variant of
construction grammar.
58 Nolan and Diedrichsen (2013) introduce a more elaborate representation
for CSs, but the crucial points distinguishing the RRG notion of construc-
tions from mainstream Construction Grammar still hold. See also
Diedrichsen (2011).
59 The undergoer can be omitted in (121a′, b′), in which case the remaining
RP is interpreted as the actor. See Van Valin (2005: 117, fn. 16), for
discussion of this aspect of the construction.
60 In Chapter 19 a way is proposed to handle languages like German and
Croatian using a new kind of syntactic template which has both fixed
and unordered features.
61 This example is adapted from Toratani (2002). The LS for taberu ‘eat’ is the
activity LS, not the full active accomplishment LS that the translation
calls for; this does not affect the points at hand.
62 This leads to a second motivation for not assigning the argument intro-
duced by -(r)are a macrorole. By the AUH in (99) it should be an actor,
which seems odd for the argument of an LS emphasizing how the
referent is affected by some event/state of affairs. If it were undergoer,
on the other hand, the semantics would be appropriate, but then a
sentence like (125b) would have two undergoers (Taroo, piza) in a single
core, which the theory doesn’t allow. The best solution is simply not to
assign a macrorole to the argument in question, as the selection of this
argument as PSA is independently accounted for in the linking system.
63 Not given, but it would have the same basic information as the German
and Japanese ‘plain’ passive CSs, mutatis mutandis, for English. See Van
Valin (2005: 132) for the English passive CS.
64 In Yaqui accusative case marking and plural marking are in comple-
mentary distribution; hence no ACC marking is possible in this sentence
with a plural undergoer.

65 More specifically, (143a) has been called ‘raising to subject’ or ‘matrix-

coding as PSA’, and (143b) has been known as ‘raising to object’, ‘excep-
tional case marking’, or ‘matrix-coding as non-PSA’.
66 While teach and forget can take that-clauses, these LSs cannot be realized
as that-clauses due to the missing arguments in the embedded LSs; they
must be realized as non-subordinate core junctures. Moreover, if oper-
ators were represented, the embedded LSs would also lack tense
operators.
67 Another instance of argument identity is reflexivization, which will not
be discussed in detail in this chapter. See Bentley (2006), González
Vergara (2006, 2009, 2011), Van Valin (1990a, b), Van Valin and LaPolla
(1997, §7.5), Van Valin (2005, §§5.3, 7.5).
68 There are two simplifications in this LS. First, bring would have a com-
plex LS involving motion and possession, but this is not relevant for the
issue at hand. Second, ‘PURP’ is the abbreviation for the purposive
interclausal relation in (94d12).
69 It should be noted that in Van Valin and LaPolla (1997: 521) the same
structure as in (155b) is posited for conjunction reduction. This analysis
is abandoned in Van Valin (2005), and the structure in (155c) is proposed
for this construction in Van Valin (2005: 229).

Part Two
Topics in RRG:
Simple Sentences

2
Lexical and Grammatical
Categories in RRG
John M. Peterson
Abbreviations

ABLE ability marker OBLIG obligation

ACT active voice OPT optative
ADJP adjectival phrase POL politeness
AOR aorist PRESUM presumptive evidential
C:TEL culminative telic V2 PROX.HEARER proximate to hearer
CONT continuous V2 PSBL possibility
EMPH emphatic QUAL qualitative predication
( ‘be’)
FAC factive REAL realis
HON honorific REM remote
HS hearsay RP reference phrase
HUM human SH subject honorific
IF illocutionary force TAM tense, aspect, mood
INT interrogative THEM theme
ISC intentional state TNS tense
construction
MID middle voice V2 unit denoting aspect/
Aktionsart
N neuter, neutral VIC verbal inflectional clitic
NHUM non-human
2.1 Introduction
The present chapter deals with the status of lexical and grammatical cat-
egories in RRG, including parts-of-speech systems and grammatical

182 JOHN M. PETERSON
categories such as tense, aspect, mood (TAM) and negation, which are
referred to in RRG as ‘operators’.* With respect to lexical categories, RRG
differs from many other theories of language which assume a small, closed
set of universal parts of speech such as ‘noun’, ‘verb’ ‘adjective’, etc., and
which consider these to be the heads of corresponding syntactic categories,
namely NP, VP and AdjP. By contrast, RRG does not assume that such
categories are universal, in line with an increasing amount of typologically
oriented research questioning the universality of these categories. Instead,
RRG assumes functionally motivated, non-endocentric syntactic categories,
such as nucleus or NUC (containing the predicate), referential phrases or
RPs, and modifying phrases or MPs. Although these syntactic slots are
typically realized by verbs, nouns and adjectives/adverbs, respectively, this
is not required by the theory. Many languages, including English, allow for
non-verbal predicates, non-adjectival modifying phrases, etc., while others
show little or no evidence for lexical categories such as ‘noun’ and ‘verb’ or
allow for clausal referents, modifiers and predicates. And even those lan-
guages which do possess traditional parts of speech often show so-called
‘category squishes’, in which assignment to a particular category is at best
an arbitrary decision. Hence, categories such as NUC, RP and MP are not
universally linked to particular lexical categories per se, although individ-
ual languages will generally have language-specific restrictions as to which
elements may occupy these slots.
The present chapter also provides a description of the grammatical cat-
egories referred to in RRG as operators, that is, those categories which
‘ground’ a clause, core or nucleus and which are closely linked to finiteness,
a topic which has been receiving increasing attention in typological
research in recent years. These categories include tense, aspect, deontic
mood (‘modality’ in RRG terms), epistemic mood (‘status’), evidentials, etc.
These units all have scope over a particular level of the sentence structure
(nucleus, core, clause) and play an important role in the determination of
nexus relations in complex sentences within the theory. This also includes
categories which are primarily concerned with questions of reference, such
as definiteness and deixis, which ground the RP.
2.2 Lexical and Syntactic Categories in RRG
2.2.1 General Issues

A basic tenet of descriptive linguistics is that linguistic categories are only
valid to the extent that they are empirically justified, and the view that
there is no such thing as a universal grammatical category is gaining
acceptance.1 This holds for all categories, including lexical categories such
* I would like to thank Utz Maas for his comments on an earlier version of the present study. Needless to say, any remaining
errors and misconceptions are my own.

Lexical and Grammatical Categories in RRG 183
as noun, verb and adjective (to name just three) as well as the correspond-
ing endocentric syntactic categories, NP, AdjP and VP, even though
these categories – or at least the lexical categories – are generally
considered universal.
Although there are considerable differences of opinion with respect to the
‘correct’ approach to this topic, most researchers would probably agree with
an approach which views lexical categories as language-specific categories
based on feature bundles, in which ‘prototypical’ members of a particular
class possess all of the features potentially associated with the particular
class.2 For example, prototypical nouns in English would denote persons,
places or things, mark for the plural (usually with -s), be compatible with
definite and indefinite articles, etc., while less prototypical nouns, for
example, may not be compatible with plural marking or the definite article.
Problematic for this approach, however, is determining which features are
more important than others, if indeed a hierarchy can be established at all,
as well as how many features are ‘enough’ for an item to be assigned to a
particular category. We will return to this topic in the following pages.
As Himmelmann (2008: 260) notes, there is little doubt that ‘the inventory
of function words is highly language-specific’. Rather, ‘[w]henever there is a
controversy regarding the number and kind of syntactic categories in a
given language, it pertains to the linguistic classification of content words’.
Himmelmann assumes two grammatical levels for the discussion of parts of
speech: (i) ‘the level of terminal syntactic categories where lexical items are
categorised according to their phrase-structural properties’, and (ii) ‘the
level of lexical categories proper where lexical items are categorised
according to those grammatical features which are not directly relevant
for phrase structure’ (Himmelmann, 2008: 263). To illustrate these two
levels,3 Himmelmann proposes a hypothetical example in a language L, in
which the most basic phrasal categories always consist of an overt function
word X and an overt content word Y, with virtually all content words being
compatible with all function words. This can be schematically portrayed as
in (1). Here the status of the function word X distinguishes the phrase from
other phrase types, hence ‘XP’.
(1) [XP [X] [Y]]
These two levels, that is, lexical and syntactic, need not correspond to one
another directly, although this will often be the case, as, for example, with
Ns as heads of NPs, Adjs as heads of AdjPs and Vs as heads of VPs in many
languages. Himmelmann (2008: 264) notes five logical possibilities with
respect to these two levels and their correspondence to one another, given
in Table 2.1. With respect to category I, both levels possess distinct categories
but these do not directly correlate in Ia (‘6¼’), while they do in Ib (‘¼’).
Returning to our example above, if, following Himmelmann, we imagine
that L has a further restriction such that not all content words Y behave
similarly, as some Ys take plural suffixes, others take plural prefixes, while

Table 2.1 Possible alignments between lexical and syntactic categories

(Himmelmann, 2008: 264)
Lexical categories Terminal syntactic categories
Ia distinct 6¼ distinct
Ib distinct ¼ distinct
II indistinct distinct
III distinct indistinct
IV indistinct indistinct
the members of a third class do not allow pluralization, then we would have
case Ia, with distinct lexical categories based on the compatibility of Y with
plural markers and also distinct terminal syntactic categories, but with no
direct correspondence between the two levels, as the compatibility of plural
affixes does not correlate with the compatibility of a particular Y with a
particular X.
To these two descriptive levels we can add a third level, that of function,
here primarily reference, predication and attribution. This can involve a
direct correlation between terminal syntactic cateogies and functional cat-
egories, that is, XP1 ¼ reference, XP2 ¼ predication, etc., or there can be no
direct correlation, so that XP1, XP2, etc. can both be used in reference,
predication and modification. At one extreme we would then have endo-
centric categories specialized for certain sentence-level functions, for
example N ¼ NP ¼ reference, V ¼ VP ¼ predication, etc., and at the other
extreme – at least putatively – no correlations whatsoever, although it is
questionable whether such a language is possible.
Such mismatches are not only a theoretical possibility; many research-
ers have claimed that the languages they are describing either do not in
fact possess a certain category such as ‘N’ or ‘NP’, or that the respective
language can at least best be described without recourse to such categor-
ies. Even in languages such as English, which possess relatively clearly
defined lexical classes such as noun, verb, and adjective, it has long been
known that there is no one-to-one relationship between part of speech
and function. Consider the examples in (2a–g), from Van Valin
(2008: 165).
(2) a. Chris will [nuc [pred see]] the movie. PRED in NUC ¼ V
b. Chris [nuc is [pred a very good detective]]. PRED in NUC ¼ NP
c. Pat [nuc is [pred exceedingly tall]]. PRED in NUC ¼ ADJP
d. Pat [nuc is [pred in the house]]. PRED in NUC ¼ PP
e. Chris [nuc [pred wiped]] PRED in NUC1 ¼ V
the table [nuc [pred squeaky clean]]. PRED in NUC2 ¼ ADJP
f. Pat [nuc [pred pushed]] PRED in NUC1 ¼ V
the table [nuc [pred out the door]]. PRED in NUC2 ¼ PP
g. Chris [nuc was [pred elected]] PRED in NUC1 ¼ V
[nuc [pred president of the club]] PRED in NUC2 ¼ NP

SENTENCE SENTENCE
CLAUSE CLAUSE
CORE CORE
RP NUC RP RP NUC
PRED AUX PRED
V NP
Chris will see the movie. Chris is a very good detective
SENTENCE
CLAUSE
CORE
RP NUC
AUX PRED
PP
Pat is in the house

Figure 2.1 Constituent projections for (2a), (2b) and (2d) (adapted from Van Valin
2008: 166)
As the examples in (2) show, the predicate of a nucleus in English need not
be verbal. While the predicate is verbal in (2a), in (2b) it is an NP, in (2c) an
AdjP and in (2d) a PP. Figure 2.1 presents the constituent projections for (2a)
(from this author), and (2b) and (2d) (adapted from Van Valin, 2008: 166).4
In English, as in many other languages, there is a general requirement
that clauses in which the predicate is not verbal must contain a kind of
‘auxiliary’ such as the copula is in order for the clause to be grammatical. As
a very good detective in (2b), exceedingly tall in (2c) and in the house in (2d) are not
verbs, the copula is is required; however, it is important to stress two things
here. First, that while is is a verb, it is not the predicate; this function is filled
by the NP in (2b), by the AdjP in (2c), and by the PP in (2d). Second, this is a
language-specific criterion which is not found in all languages or which is
only found with certain interpretations. For example, one quite common
situation is found in Russian, where a clause whose predicate is non-verbal
does not require a copula when there is a present or atemporal interpret-
ation, as in (3).
(3) Russian
eto stol
this table
‘This [is a] table.’

In other languages, the element which is often considered a copula may

not even be verbal, as in Maltese (Semitic, Malta), where the ‘copula’ is in
fact a pronoun agreeing with the subject with respect to person and number
(and gender in the third-person singular) with a present or atemporal
interpretation, as in (4).
(4) Maltese (Ambros 1998: 69)

Dan il-professur huwa Malti.
this.m.sg def-professor 3m.sg Maltese.m.sg
‘This professor is Maltese.’
As Van Valin (2008: 166) notes, non-verbal predicates do not pose a prob-
lem in RRG, since the clause is not an endocentric category: ‘Indeed, there is
no head at all. The nucleus cannot be considered the “head” of the core or
the clause, because it is not a lexical category, on the one hand, and is often
phrasal, on the other. The notion of “head” is of no relevance to the layered
structure of the clause.’ Instead, all universal categories assumed in RRG are
semantically motivated, so that in RRG a clause minimally consists of a
core – which minimally consists of a nucleus which contains the predicate –
and possibly one or more arguments.
Instead of positing universal lexical categories which are also the heads of
corresponding phrasal categories – cross-linguistically an even more prob-
lematic assumption – mainstream RRG has assumed since Van Valin (2008)
that the core contains a predicate, the functional/semantic base of the
nucleus, and possibly one or more referential phrases, referred to as RPs.
Just as the predicate need not necessarily be verbal, neither does an RP have
to be nominal. In addition, RRG assumes that these units may be modified
by modifying phrases or MPs, which can (but need not) be specialized so as
to modify only an RP, as with prototypical adjectives, or only predicative
elements, as with prototypical adverbs, although again this is not necessary,
as there are many languages with either no or only very few adjectives and/
or adverbs, where these functions are realized by other categories, such as
nouns or verbs.
Not all NPs are RPs and not all adjectives or adverbs are MPs. For
example, expletive or ‘dummy’ subjects like those in (5a) are not referen-
tial and hence not RPs: these are pronouns and therefore NPs, but they are
not referential and hence not RPs. Similarly, while a very good detective in
example (2b) is an NP, repeated here as (5b), it is not referential and
therefore also not an RP.
(5) a. It looks like it’s going to rain today.

b. Chris is a very good detective.
The same holds for the relationship between adjectives / adjective phrases
and the MP. For example, while the AdjP very good in (5b) is a modifier in the
NP a very good detective, and hence an MP, the AdjP very good in (6) is not a
modifier, and therefore not an MP.

(6) That detective is very good.
In languages which do not have the usual parts of speech such as noun,
adjective or verb, these units will of course belong to different, language-
specific categories. For example, as we will see in Section 2.2.3.1 on Kharia,
translation equivalents of nouns, adjectives and verbal stems all belong to
the open class of contentive lexemes of that language, whereas NPs and
finite verbs in English correspond most closely to the Kharia-specific units
‘Case-syntagma’ and ‘TAM/Person-syntagma’, respectively. Again, these struc-
tural units must be carefully distinguished from the functionally motivated
units RP, MP and predicate.
As mentioned above, modifying phrases or MPs in many languages have
as their default value either adjectives or adverbs, depending on what is
being modified, that is, a noun (the tall tree) or a verb (the mouse ran quickly
into the closet). These appear in the peripheries of RPs and clauses (Van Valin
2008: 172).5 As Van Valin notes, MPs also have a layered structure, for two
reasons. First, many languages allow modifier phrases with an adjectival
nucleus to take a core argument, such as the German der auf seinen Sohn
stolze Vater (lit. ‘the of his son proud father’). Secondly, the modifiers can
themselves be modified, ‘which means that they must have a periphery to
house the modifying MP, e.g. the very quickly extinguished fire’ (Van Valin
2008: 172).
In addition to adjectives and adverbs, many languages, including English,
also allow other, much more complex categories to serve as modifiers.
Consider the following examples, from the same source.
(7) a. The Charles and Di syndrome is no longer relevant.

b. The God is dead philosophers are mostly dead.
c. My grandson likes to give me the who’s the boss now, silly old grandpa wink
frequently.
In (7a) the MP contains a conjoined RP in its nucleus, that in (7b) contains

a clausal nucleus, and the nucleus of the MP in (7c) contains a constituent
question as well as a vocative in its post-detached position [PoDP]. Figure 2.2
gives the constituent projection of (7c) (PrCS ¼ pre-core slot).
Even the seemingly inconspicuous category of pre-/postpositional
phrases is not without problems. As Van Valin (2008: 170) notes, these
are normally thought of as endocentric projections of a (pre- or postpos-
itional) lexical head in predicative adpositional phrases, such as the
underlined element in (8), which has a prepositional nucleus consisting
of the preposition in, which functions as a predicate and thus licenses
the object the library.
(8) Chris saw Pat in the library.
The status of this unit in (8) is markedly different from that of the under-
lined units in (9), from the same source, referred to as non-predicative PPs as

RP
CORER
PERIPHERY NUCR
MP N
COREM
NUCM
SENTENCE
CLAUSE PoDP
PrCS CORE PERIPHERY RP
RP NUC MP
AUX PRED COREM CORER
RP NUCM PERIPHERY NUCR
ADV MP MP N
the who’ s the boss now, silly old grandpa wink

Figure 2.2 Structure of RP with MP containing sentential nucleus (adapted from Van
Valin 2008: 173)
these units do not predicate but are licensed by the verb and therefore do
not license an object.
(9) a. Chris showed the photo to Pat.

b. Chris stole the photo from Pat.
c. Chris presented Pat with the photo.
As the underlined units in (9) are non-predicative, they do not contain a

PRED and hence neither a NUC p (nucleus of the PP) nor a CORE p . Rather,
they consist merely of a pre- or postpositon and an RP. Figure 2.3 contrasts
the internal structure of these two PP types (see also Chapter 10 for adposi-
tional phrases in RRG).
With respect to these two PP types, Van Valin (2008: 171) notes that
predicative PPs most closely correspond to the X-bar notion of an endo-
centric category, as the nucleus here is always a P and always a head. He
also notes that a modifier like right in right under the table can be handled in
terms of a modifier in the periphery within the predicative PP, while with a
non-predicative PP there is no prepositional nucleus, the only nucleus being
that of the RP. As such, they can arguably be analysed as ‘exocentric PPs,

a. PP b. PP
COREP
P RP NUCP RP
CORER PRED CORER
NUCR P NUCR
N N
to the girl in the library

Figure 2.3 Non-predicative (a) and predicative (b) prepositional phrases in English (Van Valin
2008: 171)
i.e. PPs without a prepositional nucleus. Hence there is no layered structure,

and this predicts that the admittedly limited set of PP-internal modifiers
should not occur with a non-predicative PP, which seems to be the case:
*Chris showed the photo right to Pat/*Chris stole the photo right from Pat/*Chris
presented Pat right with the photo.’
2.2.2 Gradience in Lexical Categories

Any definition of lexical classes in terms of feature bundles will eventually
run into problems as there will always be at least some content morphemes
which show many but not all of the ‘prototypical’ features associated with a
certain lexical category. For example, although most nouns in English and
many other languages can appear in the plural, this is problematic with
others, such as peace. Similarly, while most nouns in German are compatible
with definite and indefinite articles, such as der/ein Mann ‘the/a man’, others,
such as Verlaub ‘leave; permission’ are compatible with neither, in this case
being restricted to occurrence in the fixed (prepositional) phrase mit Verlaub
with the preposition mit ‘with’ and the meaning ‘with respect; if you will
pardon my saying so’. Although Verlaub is listed in many dictionaries with
the definite article der (masculine, singular), it is in fact probably never
found with this article in actual speech, although it is considered a noun
for other reasons, for example its compatibility with the preposition mit,
which presupposes an NP (which is assumed to have a nominal head!) and its
perceived inherent (masculine) gender, as inherent gender is restricted to
nouns in German.
Similar problems are encountered with what Ross (1972) refers to as
‘category squishes’ or sequences of forms whose feature bundles gradually
change so that it is often difficult to decide which category they should
belong to. Van Valin (2008: 174) cites the examples in (10).

(10) a. The savant quickly proved the theorem

b. That the savant quickly proved the theorem (astonished the professors)
c. the savant/*the quickly proving the theorem (astonished the professors)
c′. the savant having quickly proved the theorem
c′′. (I am aware of ) the theorem having quickly been
proved by the savant.
d. the savant’s/*the quickly proving the theorem (astonished the professors)
d′. the savant’s having quickly proved the theorem
d′′. the theorem’s quickly being proved by the savant
e. the savant’s/the quick proving of the theorem (astonished the professors)
e′. *the savant’s quick having proved of the theorem
e′′. the quick proving of the theorem by the savant
e′′′. *the theorem’s quick proving by the savant
f. the savant’s/the quick proof of the theorem (astonished the professors)
f′. the quick proof of the theorem by the savant
f′′. the savant’s/the quick proof (astonished the professors)
Although proved in example (10a) is clearly verbal and proof in the (10f)
examples is clearly nominal, the status of the forms ending in -ing is much
less clear. While some instances of proving, such as those in (10c) and (10d),
take a direct object, suggesting that this form is a verb, the forms in (10e) and
(10e′′) require an ‘object’ marked by the preposition of, suggesting that this is
a noun. Furthermore, the ‘subject’ NP in (10c) appears as a bare NP, while in
(10d) – as in the more ‘nominal’ form in (10e) or the clearly nominal form
proof in the (10f) examples – it appears in the genitive. These and other factors
such as adjectival modification, optional complements, tense, aspect and
voice are summarized in tabular form in Van Valin (2008: 175).6
As we see in (10), even in languages such as English, which otherwise seem
to have quite clear lexical classes, it is often not feasible to assume that each
and every word form unambiguously belongs to a particular lexical
category. RRG thus assumes that it is the individual features which are
relevant and not the labels themselves:
What is relevant to the grammar is the features themselves, not any

category labels that might be overlaid on them. So ‘noun’ is just a useful
descriptive label for a certain pattern of lexical item distributions in a
language, just like ‘subject’ is a useful descriptive label for a certain
consistent pattern of restrictive neutralizations in a language.
Analogous to grammatical relations, some languages have well-defined
and sharply differentiated lexical categories, while others do not.
(Van Valin 2008: 176)
In the following two subsections, we discuss concrete examples from two

other languages, Kharia (Munda, India) and Wari’ (Chapakuran, Brazil and
Bolivia), where assuming clear-cut lexical classes and strictly endocentric
syntactic categories is even more difficult, and we will show how these
‘problematic’ elements are in fact dealt with straightforwardly in RRG.

Lexicon
Open class Closed class

(contentive morphemes,
may be used in all functions) Proforms / Grammatical
Deictics Morphemes
(referential and predicative (never used
function, used attributively referentially,
only if marked by the predicatively, or
genitive) attributively)
Figure 2.4 The Kharia lexicon (Peterson, 2011a: 78)
2.2.3 ‘Problematic’ Categories

2.2.3.1 Parts of Speech in Kharia
In the South Munda language Kharia, spoken in eastern-central India, Peter-
son (e.g. 2011a, b, 2013) argues that the familiar lexical categories of ‘noun’,
‘verb’, ‘adjective’ etc. are not only unnecessary for an adequate grammatical
description of the language, assuming their presence in fact considerably
complicates the discussion.
Peterson divides the Kharia lexicon into two major classes: the vast major-
ity of content morphemes, that is, morphemes denoting referents, attri-
butes and events/states, may be used freely in referential, attributive and
predicative functions with no overt derivational marking, ‘light verbs’, etc.
This class is open as new morphemes (e.g. modern technology and loan-
words in general) are constantly being added to its ranks. There is also a
closed class which can be further divided into two subclasses: ‘Proforms/
Deictics’, which may be used freely in referential and predicative function,
but may only be used attributively if marked for the genitive, and ‘gram-
matical morphemes’, which may never be used referentially, predicatively
or attributively. This is illustrated in Figure 2.4.
(11)–(12) present two examples demonstrating the flexibility of content
morphemes.
(11) Kharia ((11a) from Peterson (2011a: 76); (11b) adapted from Malhotra (1982: 136))
a. lebu ɖel¼ki.
man come¼mid.pst
‘The/a man came.’
b. bhagwan lebu¼ki ro ɖel¼ki.
God man¼mid.pst and come¼mid.pst
‘God became man [¼ Jesus] and came [to earth].’
(12) Kharia (Peterson 2013: 133)

[In a play about me and you, in which both of us will be taking part.]
‘naʈak¼te iɲ¼ga ho¼kaɽ¼na¼iɲ ro am¼ga iɲ¼na¼m.’ ‘umboʔ.
play¼obl 1sg¼foc that¼sg.hum¼mid.irr¼1sg and 2sg¼foc 1sg¼mid.irr¼2sg no

am¼na um¼iɲ pal¼e. ɖirekʈar seŋ¼gaʔ iɲ¼te ho¼kaɽ¼oʔ.

2sg¼inf neg¼1sg be.able¼act.irr director early¼foc 1sg¼obl that¼sg.hum¼act.pst
am¼ga am¼na¼m.’
2sg¼foc 2sg¼mid.irr¼2sg
‘“In the play I will be him and you will be me.” “No. I can’t be you. The director already
made me him. You will be you.”’
This process is entirely productive, given a proper context. See examples

(13)–(22), from Peterson (2011a: 75–92), which give some idea of the perva-
siveness of this flexibility throughout the lexicon.
(13) Interrogatives: i ‘what; which; do what?’
(14) Indefinites: jaha~ ‘something; some (attribution); do something’
(15) Quantifiers: moɲ ‘one (referential/attributive); become one’
(16) Properties: rusuŋ ‘red (one); red (attributive); become red’, maha ‘big (one);
big; grow, become big’
(17) Proper names: aʔghrom ‘Aghrom (name of a town) (referential/attributive);

come to be called “Aghrom” (middle voice), name [something] “Aghrom”
(active voice)’
(18) Status and Role: ayo ‘mother; become a mother (middle voice), accept
someone as a mother (active voice)’
(19) Deictics and proforms: iɖaʔ ‘yesterday; become yesterday (middle), turn (e.g.
today) into yesterday (active, e.g. with God as subject)’
(20) Physical objects and animate entities: kaɖoŋ ‘fish; become a fish (middle),
turn into a fish (active)’
(21) Locative: tobluŋ ‘top, rise (middle), raise (active)’
(22) Activities: siloʔ ‘plowing (n.); plowed; plow’
In addition to this precategoriality, what appear to be entire NPs can also

denote an event. Consider examples (23)–(24), both of which have the same
propositional content.

ho rochoʔb¼te col¼ki¼ɲ
that side¼obl go¼mid.pst¼1sg
‘I went to that side’

ho rochoʔb¼ki¼ɲ
that side¼mid.pst¼1sg
‘I went to that side’ (lit.: ‘I that-side-d’)
Example (24) shows that an analysis of Kharia as possessing an N/V distinc-

tion is problematic. Even if we were to treat rochoʔb ‘side’ in (24) as a
(presumably zero-derived) verb, we would then have a verb modified by
the demonstrative ho ‘that’. As (25)–(26) show, the construction in (24), with

an apparent NP as the semantic base of the predicate, is entirely productive

and can also contain both quantifiers and genitive attributes, in addition to
demonstratives. This is especially problematic for assuming ‘rampant zero
conversion’ (Evans and Osada 2005), as this would entail productive zero
derivation of a verb (i.e. a lexical stem), from a full-fledged NP (i.e. from a
complex syntactic unit).

ubar rochoʔb¼ki¼ɲ
two side¼mid.pst¼1sg
‘I moved to both sides (i.e. this way and then that).’

a. oʔ¼yaʔ teloŋ¼ki.
house¼gen roof¼mid.pst
‘The house’s roof was thatched.’
b. oʔ¼yaʔ teloŋ¼oʔ¼ki.
house¼gen roof¼act.pst¼pl
‘They thatched the house’s roof.’
On the basis of data such as these, Peterson (2013) assumes two structur-
ally defined categories, the Case-syntagma and the TAM/Person-syntagma,
both of which have the same potential structure for the semantic base and
which differ only with respect to their functional marking, namely case or
TAM/voice and person/number/honorific marking, motivated by similar
structures assumed for Tongan by Broschart (1997). The structure of the
semantic base is given in (27). None of these elements is obligatory, as long
as some non-enclitic unit is present. The Kleene star (*) denotes that poten-
tially any number of content morphemes (lex) is possible, including zero.
(27) (lex¼gen) (dem) (quant (clf )) (lex¼gen) (lex*) (¼poss) (¼num)
The maximal potential structure of the Case-syntagma is given in (28),

where X is equal to the structure of the semantic base in (27).
(28) X¼case
Case in (28) refers to postpositions such as buŋ ‘with; ins’, seŋ ‘before’, etc.,
as well as the two cases given in (29), so that any postposition, ¼te, or the
lack of any overt marking signals that the unit is a Case-syntagma.
(29) Case: Direct (zero marking)7 the case of subjects and indefinite direct objects;
Oblique (marked by ¼te) marks definite direct objects, ‘indirect
objects’ and adverbials
The genitive is not a case in this sense: whereas the direct and oblique
cases are relevant at the clause level to mark the relation of a constituent to
the predicate (roughly: subject/non-subject), the genitive is only relevant
within the semantic base and serves to integrate one semantic base into a
larger semantic base. The oblique case also cannot appear in a TAM/Person-
syntagma, as (30) shows, whereas the genitive can (31).


*sahar¼te¼ki¼ɲ.
city¼obl¼mid.pst¼1sg
‘I went to the city.’

ayo¼yaʔ¼yoʔ
mother¼gen¼act.pst.3sg
‘he or she made [it] mother’s’ (lit. ‘he or she mother’s-ed [it]’)
The structure of the TAM/Person-syntagma is given in (32), where X again

refers to the semantic base given in (27). (V2) or ‘vector verbs’, as they are
often referred to in South Asian studies, refers to phonological and morpho-
syntactic words which mark Aktionsart or the passive. voice refers to basic
voice, that is, the active and middle (Klaiman 1991).8
(32) X (V2*) (¼prf )¼tam/voice¼person/num/hon
Examples (33)–(36) illustrate this structure with a number of TAM/Person-

syntagmas in Kharia. The semantic base is underlined.

col¼ki¼may
go¼mid.pst¼3pl
‘they went’

boksel¼nom goʔɖ¼ki
sister.in.law¼2poss c:tel¼mid.pst
‘she became your sister-in-law’ (¼ ‘she your sister-in-law-ed’)

ho¼jeʔ u¼jeʔ¼ki goʔɖ¼ki
that¼sg.nhum this¼sg.nhum¼pl c:tel¼mid.pst
‘that became these’ (¼ ‘that these-d’)
(36) Kharia (AK, 1: 69)9

ro tama am¼pe u naw
~ kuʈum¼te¼ga
and now 2¼2pl this nine family¼obl¼foc
sadi biha kersoŋ¼na¼pe . . .
marry marry marry¼mid.irr¼2pl
‘And now you will marry in only these nine families . . .’
Examples (37)–(38) show a Case-syntagma and a TAM/Person-syntagma with

the same semantic base (underlined), differing only with respect to their
functional marking. Their basic structures are illustrated in Figure 2.5.10
(37) Kharia (elicited)

bides¼aʔ lebu¼ki¼yaʔ rupraŋ¼te
abroad¼gen person¼pl¼gen appearance¼obl
‘the appearance of foreigners’ (e.g. as the object of the predicate)

CASE-SYNTAGMA
SEMANTIC BASE CASE
bidesa lebukiya rupra =te
TAM/PERSON-SYNTAGMA
SEMANTIC BASE TAM/PERSON-MARKING
bidesa lebukiya rupra =ki=may.

Figure 2.5 TAM/Person- and Case-syntagmas with the same semantic base

bides¼aʔ lebu¼ki¼yaʔ rupraŋ¼ki¼may.
abroad¼gen person¼pl¼gen appearance¼mid.pst¼3pl
‘They took on the appearance of foreigners (e.g. by living abroad so long).’
These two syntactic units are defined purely structurally, and both can
appear in attributive, predicative and referential functions, as examples
(39)–(42) from Peterson (2013: 137) show.
Case-syntagma in attributive function
(39) Kharia
kuda koloŋ daru
millet bread Tree
‘a millet bread tree’ (name of a children’s story)
(Note: kuda modifies koloŋ and the two together modify daru.)11
TAM/Person-syntagma in attributive function
(40) Kharia (Peterson 2011a: 74)

yo¼yoʔj lebu col¼ki.
see¼act.pst.1sg man go¼mid.pst
‘The man I saw left.’
Case-syntagma in predicative function
(41) Kharia (AK 1: 57)

. . . ro u¼ga ho jinis¼aʔ komaŋ.
and this¼foc that animal¼gen meat
‘. . . and this [is] that animal’s meat.’

TAM/Person-syntagma in referential function
(42) Kharia (MT, 1: 180)

kunɖab aw¼ki tomliŋ khaɽiya gam ɖom¼na
behind qual¼mid.pst milk Kharia say pass¼inf
laʔ¼ki¼may ina no
ipfv¼mid.pst¼3pl because
u¼ki tomliŋ uʔɖ¼ga ɖel¼ki¼may.

this¼pl milk drink¼foc come¼mid.pst¼3pl
‘[Those who] were in the rear (¼ literally: ‘they were behind’) were called ‘Milk
Kharia’ because they came drinking milk.’
Finally, an analysis of the forms given in the preceding pages as consisting

of nominals and ‘hidden verbs’ is also not tenable. As noted above, it is not
possible in Kharia to combine TAM/Person marking with case marking or
with postpositions, so that a construction such as the one in (43), which ends
in a postposition, is not grammatical if directly followed by TAM/Person-
marking, as (44) shows.12

ikuʔɖ jughay duʔkho buŋ
very much sorrow ins
‘very sad’, literally: ‘with very much sorrow’

*ikuʔɖ jughay duʔkho buŋ goʔɖ¼ki.
very much sorrow ins c:tel¼mid.pst
‘(He) became very sad’ (¼ ‘with very much sorrow’)
However, if the semantic part of this unit is re-ordered as in (45), with two
constituents instead of just the one in (43), the unit which is not marked by
a postposition may serve as the semantic base of a TAM/Person-syntagma.
This is a common predicate type in Kharia (Peterson 2011a: 220).
(45) Kharia (RD, 1: 18)

gupa lebu duʔkho buŋ ikuʔɖ jughay goʔɖ¼ki.
watch person sorrow ins very much c:tel¼mid.pst
‘The shepherd (¼ watch person) became very sad.’
At issue here is that the unit ending in the postposition may not be directly
followed by a TAM marker, whereas the unit which does not end in a postpos-
ition/case marker can. Furthermore, if a ‘light verb’ such as hoy ‘become’
(borrowed from Kharia’s Indo-Aryan neighbour Sadri) is inserted between buŋ
‘ins’ and goʔɖ¼ki in (44), this results in a grammatical predicate, consisting of
a Case-syntagma and a marker of qualitative predication ( copula).13
(46) Kharia (Peterson 2011a: 111)

ikuʔɖ jughay duʔkho buŋ hoy goʔɖ¼ki.
very much sorrow ins become c:tel¼mid.pst
‘(He) became very sad.’

Thus ‘hidden verbs’ in Kharia would have a very different distribution

from the ‘light verb’ hoy ‘become’ or in fact any other supposed verb.
They could only be assumed to occur after ‘predicate nominals’ which are
not case-marked (e.g. for the structure in (34), i.e. those units we consider
to be the semantic base of either a TAM/Person- or a Case-syntagma), but
not in structures such as (44), so that they would also not be able to occur
where the overt ‘light verb’ hoy does (compare (44) with (46)). Thus,
hidden verbs do not simplify the analysis but rather complicate it consid-
erably, as they would not only be non-overt but would also have a unique
distribution.
Example (47) presents a simplified example of an intransitive sentence in
Kharia and Figure 2.6 presents its constituent projection in the present
analysis.
(47) Kharia (adapted from Roy and Roy 1937: 180–181)

u¼kiyar taŋ el¼aʔ sori¼ga ɲog¼e uɖ¼e¼kiyar.
this¼du now 1pl.excl¼gen com¼foc eat¼act.irr drink¼act.irr¼du
‘They two will now eat and drink with us.’
RRG can thus easily account for analyses of individual languages which do
not have supposedly universal categories such as ‘noun’, ‘adjective’ and/or
‘verb’, without needing any empty categories or hidden verbs.
Finally, there is a rather marginal type of TAM/Person-syntagma in Kharia,
quotatives, in which the semantic base consists of an entire sentence, as in
example (48).
SENTENCE
PERIPHERY CLAUSE
RP PERIPHERY CORE
CASE- TAM/PERSON-
SYNTAGMA SYNTAGMA
CORER MP PP NUC NUC RP
CORE P
CASE-
SYNTAGMA RP NUCP
NUCR COREM PREDP
NUCM PRO P PRO
u=kiyar ta el=a sori=ga og=e u =e =kiyar.
Figure 2.6 The constituent projection of (47)

(48) Kharia (Kerkeʈʈā 1990: 31)

iɖib tunboʔ ‘kersoŋ¼e la! kersoŋe la!’ loʔ¼na¼kiyar.
night daytime marry¼act.irr voc rep cont¼mid.irr¼du
‘Day and night they both will keep on [say]ing “Marry! Marry!”’
Although Peterson (2013) offers no final analysis of this form, as it is unique

in his corpus, the next section, which deals with very similar predicates in
the Chapakuran language Wari’ in which these structures appear to be
much more common, offers a possible solution to analysing predicates of
this type in Kharia.
2.2.3.2 The ‘Intentional State Constructions’ in Wari’

A similar construction to the rather marginal Kharia quotative predicate
discussed in the previous section is the much more productive ‘intentional
state construction’ (ISC) in Wari’, a member of the Chapakuran family of
Brazil and Bolivia, to which we now turn.
As Everett (2008: 383) notes, in Wari’, a predicate-initial language (VOS), in
verb-initial sentences the verb is immediately followed by the VIC or ‘verbal
inflectional clitic’, which marks for tense, voice, person, number and gender
and agrees with both subject and object. See examples (49)–(50), where the
VIC is underlined.
(49) Wari’ (Everett 2008: 383)

Quep na-in xirim te pane ta.
do 3sg.real.pst/prs-3n house father.1sg rem.pst emph
‘My father made a house long ago.’
(50) Wari’ (Everett 2008: 384)

Ten ta wao’.
weave pass.3sg type of basket
‘Baskets are woven.’
In sentences in which the verb is not the first sentence constituent, for
example with constituent questions, this sentence-initial unit is followed
by a different class of clitics which mark tense and which agree with the
gender of the item in sentence-initial position, regardless of its function in
the sentence, as in (51).
(51) Wari’ (Everett 2008: 385)

Ma’ co tomi’ na?
that.prox.hearer m/f.real.pst/prs speak 3sg.real.pst/prs
‘Who is speaking?’
As Everett (2008: 386) writes, ‘Many Amazonian languages report on

others’ thoughts, character, reactions, and other results of intentional states
by means of quotatives, i.e. literally putting words in people’s mouths.’
Wari’ also makes use of quotatives for these purposes, but their use is
apparently much more common in that language, having also spread to
other uses. What is interesting is that in Wari’, similar to the Kharia

example in (48), there is no morpheme denoting ‘say’. (52) provides an

example of this construction in Wari’, where the indices i, j and k show
the respective congruence relationships.
(52) Wari’ (Everett 2008: 386)

Ma’ co mao na -ini
that.prox.hearer m/f.real.pst/prs go.sg 3sg.real.pst/prs-3n
Guajarái naj -namk ‘oro narimak’ taramaxiconj.

Guajará (Brazilian city) 3sg.real.pst/prs-3pl.f collective woman chief
‘“Who went to Guajará?” (said) the chief to the women.’
Everett (2008: 391–392) notes a number of similarities between the ISC

predicator and simple verbs, suggesting that the clausal semantic base of
the ISC occupies the same functional slot as verbs otherwise do, namely the
predicate, although it consists of an entire clause. These include:14
• the ISC predicate appears in the clausal position otherwise occupied

exclusively by the verb
• the last syllable of the ISC predicate is stressed, as with the verb in other
sentence types. This suggests that the ISC predicate is considered a
single unit
• the predicate of an ISC may be modified just like a verb
• the predicate of an ISC may undergo compounding, like all verbs.
If the ISC predicate is considered the predicate, just like verbs in verb-initial
sentences, the following generalizations may be made for constituent order
in Wari’ (Everett 2008: 392):
• Wari’ sentences always begin with a verb, ISC predicate, or preverbal

mood marker
• the VIC appears directly after the predicate, whether this is clausal or a
simple verb
• tense appears in sentence-second position.
Everett (2008: 399) argues that Wari’ inflectional clitics – that is, those
which follow the predicate of the clause – follow the NUC in an RRG
analysis. As such, they follow the ISC predicate as well, which then
occupies the NUC slot. Figure 2.7, adapted from Everett (2008: 398),15
shows the structural analysis which he assumes for example (52), where
the entire ‘quote’ is viewed as a nucleus.16 The structure in Figure 2.7
captures the fact that the nucleus of the clause, which contains the
predicate, consists of a form which, despite its role as the predicate, also
has many of the trappings of a clause, as it consists of a precore unit and
a core, which itself contains a nucleus (predicate) and two arguments.
Nevertheless, it is clearly the predicate of the larger clause. This then,
together with the two RPs narima’ and taramaxicon, forms the core of the
higher-level clause.

SENTENCE
CLAUSE
CORE
NUC RP RP
PRED
CLAUSE
PreCore CORE
NUC PRO RP
NP PRED
Ma’ co mao na-in Guajará nanam ‘oro narima’ taramaxicon

Figure 2.7 Constituent projection of (52) (adapted from Everett 2008: 398)
2.3 Grammatical Categories: The Operators
A great deal of research has traditionally gone into modelling an appropri-

ate representation of the argument structure of predicates – both in RRG
and other syntactic theories. As discussed in Chapter 6 of the present
volume, in RRG this information is stored in the lexicon in the form of
logical structures, which contain information on the basic meaning of the
predicate, its Aktionsart class(es), and the number of arguments and their
thematic relations. The predicate and its arguments provide us with what
can be termed the ‘sentence base’ (see Maas 2004), that is, the basic propos-
itional information, but without any information relating to the specifics of
the utterance (i.e. is the information a command, a request, a statement or a
question? Does it refer to a past, present, future, iterative or hypothetical
event/state? Are the referents identifiable, etc.?) Information of this type is
referred to as ‘sentence modality’ or ‘semantic finiteness’ in Maas (2004), to
distinguish it from the overt marking of this information, which he refers to
as ‘morphological finiteness’. This information is also often referred to
collectively as the ‘grounding’ or ‘anchoring’ of the sentence. This can be
portrayed as in Figure 2.8.
This grounding corresponds closely to Tomasello’s (2008) three kinds of
‘syntax’ , that is, ‘simple’, ‘serious’ and ‘fancy syntax’, which accompany
what he views as the three basic evolutionary foundations of human
cooperative communication:

Sentence
Sentence base Sentence modality
Predicate A1, A2, A3, … Speech act type Time Questions of reference …
(command / of referential
request, declarative, state / identifiable
question, …) event …
Figure 2.8 The sentence base and sentence modality
Thus, since requesting prototypically involves only you and me in the

here and now and the action I want you to perform, combinations of
natural gestures and/or linguistic conventions require no real syntactic
marking but only a kind of ‘simple syntax’ in a grammar of
requesting . . . But when we produce utterances designed to inform
others of things helpfully, this often involves all kinds of events and
participants displaced in time and space, and this creates functional
pressure for doing such things as marking participant roles and speech
act functions with ‘serious syntax’ in a grammar of informing. Finally,
when we want to share with others in the narrative mode about a
complex series of events with multiple participants playing different
roles in different events, we need even more complex syntactic devices
to relate the events to one another and to track the participant across
them, which leads to the conventionalization of ‘fancy syntax’ in a
grammar of sharing and narrative.
(Tomasello 2008: 244–245).17
In RRG, this information is divided into two broad categories: the first group
we will refer to as ‘proposition-grounding’, as these categories ground the
event or state with respect to time, modality, speech act, etc. The second
group we will refer to as ‘referent-grounding’, as it refers to
referential identifiability.
2.3.1 Proposition-Grounding Operators

We begin with a discussion of propositional grounding, which is expressed in
individual languages by language-specific grammatical categories, such as
present tense, past tense, hearsay, mirative, etc. The grounding information
expressed by these language-specific categories is divided into three different
groups according to which level of the layered structure of the clause (LSC)
they ground, the nucleus, the core or the clause. RRG assumes that individual
languages will differ as to which of these grounding concepts are expressed
through grammatical categories, and there is no requirement that every
language have all of the possible operators. In fact, most languages do not
have grammatical categories expressing all of these semantic categories,

Table 2.2 Operators in the layered structure of the clause, from Van Valin (2005: 9)
Nuclear operators:
Aspect
Negation
Directionals (only those modifying the orientation of an action or event without reference to the
participants)
Core operators:
Directionals (only those expressing the orientation or motion of one participant with reference to
another participant or to the speaker)
Modality (root modals, e.g. ability, permission, obligation)
Internal (narrow scope) negation
Clause operators:
Tense
Evidentials
Illocutionary force (IF)
although it is assumed that all languages will have grammatical means to

express at least two of these categories – negation and illocutionary force (Van
Valin 2005: 9). These operators are summarized in Table 2.2.
Nuclear operators such as aspect or directionals modify the state/event
itself without reference to the participants, while the core-level operators
modify the relation between the arguments and the event, most notably
directionals, (deontic) modality (‘be able’, ‘must’, etc.), narrow scope neg-
ation, and event quantification. Finally, operators which have scope over the
entire clause include ‘status’ (epistemic modality and external negation), as
well as tense, evidentials and illocutionary force.
Operators are represented by an operator projection, given below the
respective language example. This projection is joined to the constituent
projection through the nucleus. Figures 2.9 and 2.10 illustrate this for
English and Japanese.
Note that in the left-hand example in Figure 2.9 the information for the
aspect operator comes from two different positions: the auxiliary be preced-
ing leaving and the suffix -ing on this form. In fact, this information can
potentially appear anywhere in the sentence. The only claim that RRG
makes with respect to the placement of operators is that ‘the ordering of
the morphemes expressing operators with respect to the verb indicates their
relative scopes. That is, taking the nucleus as the reference point, the
morphemes realizing nuclear operators should be closer to the nucleus than
those expressing core operators, and those manifesting clausal operators
should be outside of those signaling nuclear and core operators’ (Van Valin
2005: 11). All evidence so far suggests that this is true of language in general.
Although there may be language-specific variation of the ordering of the
operators within a particular level, cross-linguistic data for operator
marking appears to substantiate this claim for those markers for different
levels which appear together on the same side of the nucleus: see examples
(53)–(57) (from Van Valin 2005: 10).18

SENTENCE SENTENCE
CLAUSE CLAUSE
CORE CORE
RP NUC RP NUC
PRED AUX PRED
V PP
Will they have to be leav-ing? Pat was in the library.
V PP
ASP NUC NUC
MOD CORE CORE
TNS CLAUSE TNS CLAUSE
IF CLAUSE IF CLAUSE
SENTENCE SENTENCE
Figure 2.9 Constituent and operator projections for two English sentences (adapted from
Van Valin 2005: 14)19
SENTENCE
CLAUSE
PERIPHERY CORE
RP RP NUC
PRED
PP V
Taroo ga Kazue no uti de hon o yon-da ka?

NOM GEN house in book ACC read-PST Q
‘Did Taroo read a/the book at V

Kazue’s house?’
NUC
CORE
CLAUSE TNS
CLAUSE IF
SENTENCE
Figure 2.10 Constituent and operator projections in a Japanese sentence (adapted from
Van Valin 2005: 14)

(53) Kewa (Papua-New Guinea) (Van Valin 2005: 10)

a. Íra-pa-niaa-ru.
cook-prf-down-1sg.pst
‘I burned it downward (as a hill).’ (V-ASP-DIR-TNS)
b. Íra-pa-saa-ru.
cook-prf-up-1sg.pst
‘I burned it upward (as a hill).’ (V-ASP-DIR-TNS)
(54) Turkish (Van Valin 2005: 10)

a. Gel-iyor-du-m.
come-prog-pst-1sg
‘I was coming.’ (V-ASP-TNS)
b. Gel-emi-yebil-ir-im.
come-able.neg-psbl-aor-1sg
‘I may be unable to come.’ (V-MOD-STA-TNS)
(55) Korean (Van Valin 2005: 10)

An-tul-li-wu-(si)-lswuiss-cianh-ass-keyss-up-nita.
neg-hear-caus-pass-(sh)-able-neg-pst-presum-pol-decl
‘(I) guess that (he) [HON] might not be heard.’ (NEG-V-MOD-NEG-TNS-EVID-IF)
(56) English (Van Valin 2005: 10)

a. He may be leaving soon. (IF/TNS-STA-ASP-V)20
b. She was able to see them. (IF/TNS-MOD-V)
c. Will they have to be leaving? (IF/TNS-MOD-ASP-V)
(57) Tiwi (Van Valin 2005: 10)

ŋ -ru-untiŋ-apa.
e
¯¯
1sg-pst-prog-eat
‘I was eating.’ (TNS-ASP-V)
Cross-linguistic data on the position of operators (or markers of finite-

ness in general)21 within the sentence also suggests that these operators
can theoretically appear at any position in the sentence and are not
restricted to any particular position (e.g. adjacent to the verb, sentence-
final, or sentence-second (i.e. the Wackernagel position)), and it is only
their respective order when they appear together on the same side of the
predicate which is restricted. For example, in his cross-linguistic study of
finiteness, Maas (2004) suggests the typology of morphological finiteness
given in Table 2.3. Types I, II and III denote simple predicates which
consist of a single predicating unit (generally a simple verb), while Types
Ia, IIa, IIb and IIc are complex predicate types which consist of at least
two parts (whether verbal or not) which together express a
single predicate.
Types I and Ia are typical of isolating languages. Here, neither the simple
predicate, consisting of a single element (usually a verb) (Type I, cf. (58)a),
nor either of the two or more parts of the complex predicate (Type Ia) is
morphologically marked for ‘finite’ categories (cf. (58)b).

Table 2.3 Predicate types with respect to ‘finiteness’ marking
Type Scheme
I Eþ P No (verbal) morphology
Ia Eþ P (Modifying þ Modified) No (verbal) morphology
II Eþ P-finite Morphologically finite predicate
IIa Eþ P (Modifying þ Modified-finite) Modified element (¼ ‘lexical verb’) is finite,
modifying (or ‘situating’) element is
invariable (e.g. ‘particle’)
IIb E þ P (Modifying-finite þ Modified) Modifying element is finite, modified
element is non-finite (¼ auxiliary-
participle type)
IIc E þ P (Modifying-finite þ Modified-finite) Both elements are finite
III (E þ P)-finite Movable affixes
(Maas 2004: 379, modified and reprinted with kind permission by de Gruyter)
E refers to the complement of a predicate, which may be either an argument or an adjunct;
P denotes the predicate; þ refers to a loose concatenation of words and - to a narrow
morpheme concatenation.
(58) Mandarin Chinese

a. Type I: tā qù
3sg go
‘s/he goes’
b. Type Ia: tā yào qù
3sg want/fut go
‘s/he wants to / will go’
In Type II, typical of fusional and agglutinating languages, the simple

predicate is marked for all finite categories (Type II), while for the complex
predicates in Types IIa–c either the modified element (i.e. the lexical verb),
the modifying or situating element (e.g. the auxiliary) or both is/are marked
for all finiteness categories. Note that only complex predicates of Type IIb
correspond to the traditional notion of auxiliaries. Also, as the Maltese
examples in (59b–c) and (61a–b) show, languages can belong to more than
one category.
(59) Maltese (Semitic, Neo-Arabic, Malta)

a. Type II ji-kteb
3sg.m-write.ipfv fut
‘he writes’
b. Type IIa sa ji-kteb
fut 3sg.m-write.ipfv
‘he will write’
c. qed ji-kteb
prog 3sg.m-write.ipfv
‘he is writing’

(60) Type IIb: English

a. She has written.
b. I have gone.
(61) Type IIc: Maltese

a. ir-rid im-mur
1sg-want.ipfv 1sg-go.ipfv
‘I want to go’ (literally: ‘I-want I-go’)
b. kon-t mur-t
cop.pst-1sg go.pfv-1sg
‘I had gone’ (literally: ‘I-was I-went’)
Finally, in Type III at least some of the language-specific finiteness categor-

ies are marked elsewhere in the clause and not together with the other
finiteness categories. For example, in (62), from Paez, a Chibcha language
(possibly a language isolate) spoken in Columbia, all finiteness markers
attach to the predicate in the default case, as in example (62a), with the
form uʔxuetstxu ‘I went’, where the marker -txu ‘fac.1sg’ directly attaches to
the verb form. In (62b), on the other hand, it attaches (in its phonologically
determined alternative form -tx) to the nominal element nengasu ‘through
Belcazar’ to mark this element as focused, while the verb form uʔxuets is now
‘partially finite’ (see the discussion in Maas 2004: 378).
(62) Type III: Paez (Maas 2004: 378)

a. xuʔna ʧamb-na uʔx-ue-ts-txu juʔ nava kar:o suw-ku.
yesterday village-to go-ipfv-prog-fac.1sg them but car broken-fac.3sg
‘Yesterday I went to the village, but the car was broken.’
b. nenga-su-tx uʔx-ue-ts juʔ saʔ tja-xu paʔ jaʔ-tx.

Belcazar-through-fac.1sg go-ipfv-prog them and dem-from come already-fac.1sg
‘I passed Belcazar and am already back from there.’
As noted above, RRG does not posit a privileged position for such marking,
and markers of these categories can appear almost anywhere in the sentence.
This allows us to account for such cross-linguistic variability in a straightfor-
ward way, without having to resort to movement or other processes to derive
the actual word order in a particular language. In fact, the only claim that
RRG makes is that the ordering of these morphemes with respect to the
predicate – when they appear on the same side of this unit – corresponds to
the relative scopal properties of the individual operators. And in the case of
the Paez example in (62) this is unproblematic, as the ‘mobile’ markers are
clausal operators which appear in the final position when attached to the
verb in (62a), after aspectual marking and therefore conforming to this
constraint. The position of this marking in (62b) directly preceding the
predicate is also not an exception to this rule, as here the marking of this
clausal operator is not on the same side of the verbal root as the other
operators and hence this constraint does not apply here.

SENTENCE
CLAUSE
CORE
PRED
TAM/PERSON-syntagma
NUC
NUC NUC RP
le col kan =ki Ø Ø ‘It [= the parrot]

fly go CONT=MID.PST DECL 3SG continued flying away.’
NUC NUC
NUC ASP
CORE
CLAUSE TNS
CLAUSE IF
SENTENCE
Figure 2.11 Constituent and operator projections in Kharia
Finally, as discussed in Chapter 13 of this volume, the scopal properties of

these operators play a pivotal role in the discussion of nexus and juncture
relations in RRG. Consider Figure 2.11, from Kharia (own data), which
provides an example of nuclear cosubordination, as the scope of kan, a
marker of aspect/Aktionsart, has obligatory scope over both nuclei, leŋ ‘fly’
and col ‘go’.
Operators are represented formally in the semantic representation of the
sentence in italicized caps within angled brackets indicating their scope
within the logical structure. The overall pattern is given in (63), from Van
Valin (2005: 50), with arbitrarily chosen values. Note that each of the indi-
vidual operators refers to a grammatical category within a specific language
and as such its value is subject to the system of that language, for example,
with respect to tense, which could be pst/npst in one language, pst/prs/
fut in another, or fut/nfut in yet another.
(63) 〈IF DEC 〈EVID HS 〈TNS PST 〈STA IRR 〈NEG Ø 〈MOD OBLIG 〈EVQ SG 〈DIR Ø 〈ASP PERF
〈LS〉〉〉〉〉〉〉〉〉〉

(64), also from Van Valin (2005: 50), provides an example of the
formal representation of the operators for the English question Has Kim
been crying?.
(64) 〈IF INT 〈TNS PRS 〈ASP PERF PROG 〈do′ (Kim, [cry′ (Kim)])〉〉〉〉
2.3.2 Reference-Grounding Operators

As noted in Section 2.3, the RP also partakes in the semantic/pragmatic
grounding of the clause, for instance with respect to definiteness. Table 2.4
summarizes those operators which are assumed for the RP in RRG. Table 2.4
is still somewhat tentative as it is based on the operators which RRG
assumed for the NP before the notion of the RP was introduced in Van Valin
(2005: 28) and further justified in Van Valin (2008). We thus tentatively
assume that the same operators hold for the RP as for the NP, as we are
dealing with the same underlying function here, namely the semantic/
pragmatic grounding of referential units, although this topic requires
further study.
As with the sentence, RPs also possess an operator projection. Figure 2.12
provides an example of such a projection for an RP in English.
Similar to the position of sentence-grounding operators discussed in the
previous section, there are no general requirements in RRG that the individ-
ual operators appear adjacent to any particular element(s) of the RP, only
language-specific rules regulating this, although it is expected that oper-
ators appearing on the same side of the nucleus will be arranged according
to their scope, with nuclear operators appearing closer to the nucleus,
followed by core operators.
To give an example of the positional variability of these operators,
consider the different orders in the Dyirbal examples in (65). Dyirbal is
spoken in northern Australia and has exceptionally free word order. Note
that all three variants in (65) are grammatical and have the same meaning.
Note also the position of the determiners with respect to the nouns which
these modify.
Table 2.4 Operators in the layered structure of the RP (adapted from Van Valin
2005: 24)
NuclearR operators:
Nominal aspect (count–mass distinction, classifiers in classifier languages)
CoreR operators:
Number
Quantification (quantifiers)
Negation
RP operators:
Definiteness
Deixis

RP
CORER
PERIPHERYR NUCR
ADJ N
the three big bridge-s
NUCR
CORER NUM
QNT CORER
DEF RP
Figure 2.12 Layered structure of the RP in English (adapted from Van Valin 2005: 25)
(65) Dyirbal (Van Valin 2005: 28)

a. Bayi bargan baŋul yaɽa-ŋgu d̨ urga-ɲu gambi-ɽa.
det.abs wallaby.abs det.erg man-erg spear-tns mountains-loc
‘The man speared the wallaby in the mountains.’
b. Baŋul gambi-ɽa yaɽa-ŋgu bayi d̨ urga-ɲu bargan.
det.erg mountains-loc man.erg det.abs spear-TNS wallaby.abs
c. Yaɽa-ŋgu d̨ urga-ɲu gambi-ɽa bargan baŋul bayi.
man.erg spear-TNS mountains-loc wallaby.abs det.erg det.abs
The operator projection of the various orders in (65) is straightforward and is

given in Figure 2.13, which presents the projection of the structures in (65b).
Similar to the proposition-grounding operators in the semantic represen-
tation of the sentence, the reference-grounding operators are represented in
the semantic representation of the RP. An example is given in (66), from Van
Valin (2005: 52), assuming that these are the same for the RP as those
previously assumed for the NP. The conventions used here (i.e. angle
brackets denoting scope, italicized small caps, etc.) are the same as those
used to represent the proposition-grounding operators. 9 is the existential
operator, roughly meaning ‘there is (at least one)’.
(66) 〈 DEIC PROX 〈 DEF þ 〈 NEG Ø 〈QNT 9 〈 NUM SG 〈 nasp count 〈N〉〉〉〉〉〉〉
The logical structure for a phrase like the scarf would then be represented as
in (67), also from Van Valin (2005: 52).
(67) 〈DEF þ 〈NEG Ø 〈QNT 9 〈NUM SG 〈NASP COUNT 〈(scarf )〉〉〉〉〉〉
Although in principle every RP in a sentence has such a semantic repre-

sentation, for practical reasons it is seldom feasible to include all of this

SENTENCE
CLAUSE
PERIPHERY CORE
RPERG NUC RPABS
RPLOC CORER PRED CORER
RPERG CORER NUCR RPABS NUCR
PRODEM NUCR PRODEM
N N V N
Ba ul gambi- a ya a- gu bayi durga- u bargan.

DET.ERG mountains-LOC man-ERG DET.ABS spear-TNS wallaby.ABS
N N V N
NUCR NUCR NUC NUCR
CORER CORER CORE
RP CLAUSE TNS CORE R
DEIC RP SENTENCE
DEIC RP
Figure 2.13 The structure of the Dyirbal sentence in (65b) (adapted from Van Valin 2005: 29)
information in the full semantic representation of the sentence, hence it is

generally omitted. An example of a full structure is given in (68) for the
simple intransitive question Has the tall man been crying? (Van Valin 2005: 53).
The underlining of man in the semantic representation of the RP denotes
that it is the head of the RP.22
(68) 〈IF INT 〈TNS PRES 〈ASP PERF PROG 〈do′ (x, [cry′ (x)] 〉〉〉〉
〈DEF + 〈QNT 9 〈NUM SG 〈NASP COUNT 〈be′ (man (x), [tall′])〉〉〉〉〉

-----
2.4 Questions for Future Research
Rauh (2010: 378–383) raises a number of important questions with respect to

lexical and syntactic categories in RRG which need to be addressed and to
which we now turn. These questions can be divided into two broad categories.
The first are questions concerning the status of lexical and syntactic categor-
ies and how these relate to one another in RRG. For example, Rauh (2010: 380)
raises a question with respect to examples (2b–d) above, repeated here in (69).

(69) b. Chris [nuc is [pred a very good detective]]. PRED in NUC ¼ NP

c. Pat [nuc is [pred exceedingly tall]]. PRED in NUC ¼ ADJP
d. Pat [nuc is [pred in the house]]. PRED in NUC ¼ PP
As Rauh notes, is in these examples is not viewed as the predicate in RRG,

but rather the predicate is the NP, AdjP and PP, respectively. As such, the
question arises as to how is is introduced into the sentence. The usual
answer would be to assume that English, like many languages, requires a
‘dummy verb’ such as the copula when the predicate is not verbal. However,
as Rauh rightly notes, if labels such as noun and verb are merely useful
descriptive labels for certain language-specific feature-bundle patterns and
not universally valid categories, this is not an option.
As has been argued in the preceding pages, in RRG there is no need to
specify that predicates which are traditionally viewed as consisting of
NPs, AdjPs, PPs, etc., require a ‘dummy verb’ – in English or elsewhere.
Recall from Section 2.2 that it is the features which are relevant to the
grammar, not these convenient categorial labels (Van Valin 2008: 176). It
would therefore be more correct to say that a copula is required with
stative predication in English when the distributional properties of the
content morpheme of the predicate are not compatible with TAM/person
marking (*Pat (exceedingly) talls.). No reference is necessary to categories
such as noun, adjective, verb, etc. Rather, the theory recognizes that
individual content morphemes can have language-specific distributional
properties, which can only be determined through tests and not through
‘logical’ conjecture, based, for example, on whether they denote an action,
object, etc. These language-specific restrictions include the (non-)compati-
bility of a particular content morpheme with TAM or case markers,
determiners, etc. However, as we saw in Section 2.3.1 for Kharia, this is
not equivalent to saying that these content morphemes actually are
nouns or verbs. Also, in view of the problems with ‘category squishes’
described in Section 2.2, where only some of the supposedly defining
features of a particular category are present, direct reference to specific
features as opposed to (at best more-or-less predictable) feature bundles
provides us with a much more exact description of the grammatical
operations involved.
Another area mentioned by Rauh, qualia structure (Pustejovsky 1995), is
of a very different nature and raises more serious questions, which we can
only touch upon here. Qualia structure is one of four interpretive levels,
the other three being argument structure, event structure and the inter-
action of semantic levels. This level consists of four basic roles (Pustejovsky
1995: 76):
• CONSTITUTIVE: the relation between an object and its constituent parts

• FORMAL: that which distinguishes the object within a larger domain
• TELIC: its purpose and function
• AGENTIVE: factors involved in the origin or ‘bringing it about’.

Rauh refers to qualia structure on two occasions with respect to open

questions in RRG, highlighting what is in her view the importance of
lexical categories in constructing the logical structure of the sentence
from the lexicon. For example, ‘Lexically, the NPs should be represented
by items which are described on the basis of qualia structures in the
lexicon, since such a description identifies them as Ns and indicates their
compatibility with those peripheries and operators that are N-specific
rather than CLAUSE-specific’ (Rauh 2010: 380). Also, ‘To begin with, in
order to construct the semantic representation of predicates and their
arguments, a specification of nouns (items with a qualia structure) is
required’ (Rauh 2010: 382). As just noted, we believe that there is no need
to refer to lexical categories in describing the compatibility which Rauh is
referring to, if such restrictions are described in terms of individual fea-
tures and not lexical categories.
The status of qualia structure, however, requires some discussion. To
begin with, the presence of qualia structure is not restricted to ‘nouns’,
since verbs also have qualia structure (see e.g. §6.2.5 in Pustejovsky 1995).
Nor is event structure restricted to ‘verbs’, since nominalizations or nouns
such as war in English also have event structures (see e.g. Pustejovsky, 1995,
§§8.4 and 8.5). Of course, the structure and type of information contained at
these levels depends among other things on whether an action or an object
is being described, but it is worth recalling here that saying that a content
morpheme denotes an action is not equivalent to saying that it is a ‘verb’
with a predictable set of distributional (and other) features, and to my
knowledge, the lexical structures given in Pustejovsky (1995) do not contain
any direct reference to such lexical categories, although they do, of course,
contain information relating to semantic features, such as ‘physobj’, ‘pro-
cess’, ‘state’, etc.
This issue requires further study, and it is likely that we will have to re-
evaluate a number of assumptions with respect to semantic structure in
order to be able to do away with lexical categories entirely. Furthermore,
languages such as Kharia, discussed in Section 2.3.1, where virtually all
content morphemes can be used predicatively, referentially and attribu-
tively, will undoubtedly present us with new challenges with respect to
structuring the lexicon. However, as lexical categories do not play a very
prominent role in Pustejovsky’s (1995) formalism, this does not appear to be
an insurmountable task, although Rauh (2010: 378–383) is certainly correct
in noting that this will require assuming a much richer lexicon in RRG,
containing considerably more language-specific information on individual
morphemes than is currently the case. At any rate, languages such as Kharia
with its almost total lack of distributional restrictions on content mor-
phemes, or Wari’ with its clausal predicates in the highly productive inten-
tional state constructions, discussed in Section 2.3.2, as well as ‘category
squishes’ even in languages where we have come to accept the presence of
clear-cut lexical categories as self-evident, force us to accept that these

highly intuitive lexical categories cannot play a role in a theory of language

with claims to universality.
In sum, while much work remains to be done, it seems clear that trad-
itional ‘categorial thinking’ cannot serve as the basis for a universal theory
of language and is in fact at the heart of many descriptive problems which
have plagued linguistic theory for decades. The view taken here is that these
problems can only be overcome by the wholesale rejection of such categories
in language theory, granting them nothing more than the status of conveni-
ent labels in informal discussion.
References
Ambros, Arne A. 1998. Bonġornu, kif int? Einführung in die maltesische Sprache.
Wiesbaden: Reichert Verlag.
Bisang, Walter. 2001. Finite vs. non finite languages. In Martin Haspelmath,
Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language
Typology and Language Universals, Vol. 2, 1400–1413. Berlin: Mouton de
Gruyter.
Broschart, Jürgen. 1997. Why Tongan does it differently: Categorial distinc-
tions in a language without nouns and verbs. Linguistic Typology 1(2):
123–165.
Croft, William. 2000. Parts of speech as language universals and as language-
particular categories. In Petra M. Vogel and Bernard Comrie (eds.),
Approaches to the Typology of Word Classes (Empirical Approaches to Lan-
guage Typology, 23), 65–102. Berlin: Mouton de Gruyter.
Croft, William. 2005. Word classes, parts of speech, and syntactic argumen-
tation. Linguistic Typology 9(3): 431–441.
Evans, Nicholas and Toshiki Osada. 2005. Mundari: The myth of a language
without word classes. Linguistic Typology 9(3): 351–390.
Everett, Daniel L. 2008. Wari’ intentional state constructions. In Van Valin,
Jr. (ed.), 381–410.
Håkansson, Gisela and Jennie Westander. 2013. Communication in Humans and
Other Animals (Advances in Interaction Studies, 4). Amsterdam: John
Benjamins.
Haspelmath, Martin. 2007. Preestablished categories don’t exist: Consequences
for language description and typology. Linguistic Typology 11(1): 119–132.
Haspelmath, Martin. 2011. The indeterminacy of word segmentation and
the nature of morphology and syntax. Folia Linguistica 45(1): 31–80.
Hengeveld, Kees and Jan Rijkhoff. 2005. Mundari as a flexible language.
Linguistic Typology 9(3): 406–431.
Himmelmann, Nikolaus P. 2008. Lexical categories and voice in Tagalog. In
Peter K. Austin and Simon Musgrave (eds.), Voice and Grammatical Relations
in Austronesian Languages (Studies in Constraint-Based Lexicalism),

Hockett, Charles F. 1960. The origin of speech. Scientific American 203: 88–111.
[Reprinted in Wang, William S.-Y. 1982. Human Communication: Language
and Its Psychobiological Bases. Scientific American: 4–12].
Kerkeʈʈa, K. P. 1990. Jujhair ɖā̃ɽ (khaɽiyā nāʈak). Ranchi: Janjātı̄ya Bhāsā
˙
Akādemı̄, Bihār Sarkār.
Klaiman, M. H. 1991. Grammatical Voice (Cambridge Studies in Linguistics,
59). Cambridge: Cambridge University Press.
Luuk, Erkki. 2010. Nouns, verbs and flexibles: Implications for typologies of
word classes. Language Sciences 32(3): 349–365.
Maas, Utz. 2004. ‘Finite’ and ‘nonfinite’ from a typological perspective.
Linguistics 42(2): 359–385.
Malhotra, V. (1982). The Structure of Kharia: a Study of Linguistic Typology and
Language Change. Unpublished PhD dissertation. New Delhi: Jawaharlal
Nehru University.
Nikolaeva, Irina (ed.). 2007. Finiteness. Theoretical and Empirical Foundations.
Peterson, John. 2005. There’s a grain of truth in every ‘myth’, or, Why the
discussion of lexical classes in Mundari isn’t quite over yet. Linguistic
Typology 9(3): 391–405.
Peterson, John. 2011a. Kharia. A South Munda language (Brill’s Studies in the
Languages of South and Southwest Asia, 1). Leiden: Brill.
Peterson, John. 2011b. Aspects of Kharia grammar – A Role and Reference
Grammar (RRG) approach. In Rajendra Singh and Ghanshyam Sharma
(eds.), Annual Review of South Asian Languages and Linguistics, 81–124. Berlin:
Mouton de Gruyter.
Peterson, John. 2013. Parts of speech in Kharia: a formal account. In Rijkhoff
and van Lier (eds.), 131–168.
Pustejovsky, James J. 1995. The Generative Lexicon. Cambridge, MA: MIT Press.
Rauh, Gisa. 2010. Syntactic Categories: Their Identification and Description in
Linguistic Theories (Oxford Surveys in Syntax and Morphology, 7). Oxford:
Rijkhoff, Jan and Eva van Lier. 2013. Flexible Word Classes: Typological Studies of
Underspecified Parts of Speech. Oxford: Oxford University Press.
Ross, John Robert. 1972. The category squish: Endstation Hauptwort. In
Papers from the Eighth Regional Meeting. Chicago Linguistic Society. April 14–16,
1972, 316–328.
Roy, Sarat Chandra and Ramesh Chandra Roy. 1937. The Khāriās. Ranchi: Man
˙
in India.
Tomasello, Michael. 2008. Origins of Human Communication (The Jean Nicod
Lectures). Cambridge, MA: MIT Press.
Van Valin, Robert D., Jr. 2008. RPs and the nature of lexical and syntac-
tic categories in Role and Reference Grammar. In Van Valin (ed.),
161–178.

Pragmatics Interface (Studies in Language Companion Series 105). Amster-
dam: John Benjamins.
Notes
1 See, for example, Haspelmath (2007).

2 A notable exception to this is Croft (e.g. 2000, 2005), who approaches the
topic from a cross-linguistic perspective and hence begins with seman-
tics instead of language-specific morphosyntactic categories, assuming
prototypical nouns (¼ objects), verbs (¼ actions) and adjectives (¼ attri-
butes) and mapping language-specific forms/constructions against these.
As what is at issue for our discussion at this point is the necessity (or
rather, the lack thereof ) for assuming universally valid categories for
language-specific feature bundles, we follow here the approach laid out
in Himmelmann (2008). Further approaches to the topic abound in the
literature; to cite just a few of the more recent typological works on this
topic: Evans and Osada (2005) and the responses to this article (Peterson
2005; Croft 2005; Hengeveld and Rijkhoff 2005; Luuk 2010; and the many
contributions in Rijkhoff and van Lier 2013).
3 In the following, we will assume that there is a clear distinction between
‘word level’ and ‘syntactic level’, a prerequisite for this analysis,
although it is well known (see e.g. Haspelmath 2011) that the boundary
between the two levels is often not as clear as generally assumed.
4 The operator projections have been omitted here as these are not dis-
cussed until Section 2.3.
5 For an example of this, see the three peripheries in Figure 2.2.
6 See Maas (2004) for similar data from German (p. 368) and Portuguese
(p. 374).
7 Zero marking is restricted to the Case-syntagma, as the TAM/Person-
syntagma is always overtly marked as such. Thus, the lack of any
marking is unambiguous as it is only found with Case-syntagmas in
the function of subject or indefinite direct object.
8 As the passive is marked by a V2, not by an obligatory fusional TAM/Voice
marker, and always occurs together with the middle voice, it has a different
status than the active and middle, which can be considered ‘basic’.
9 Sources given with this format are from this author’s own corpus. The
letters are the speaker’s initials. The first number refers to the text from
this speaker and the second number, after the colon, refers to the line of
this text.
10 The triangles in Figure 5 indicate that there is further internal structure
which we cannot deal with here but which does not affect the discussion
at hand. For a fuller discussion, see Peterson (2013).

11 In Peterson (2011b, §4.6.1) it is shown that kuda koloŋ daru is not

a compound.
12 The following discussion and the corresponding examples are from
Peterson (2013: 145–146).
13 This discussion and the accompanying examples have been taken from
Peterson (2011a: 111).
14 For reasons of space, Everett’s argumentation has been somewhat
simplified here.
15 Here and in the following, minor changes will be made to analyses and
figures, such as the use of ‘RP’ as opposed to ‘NP’, as the concept of the RP
was not fully justified until Van Valin (2008). Furthermore, minor
changes with respect to glosses, etc., have silently been made and will
not be commented on in the following figures.
16 The representation of the operators has been left out of Figure 2.7, as
these are not dealt with until Section 2.3.
17 Of these three functions, the last two, i.e. helpful informing and the
narrative mode, seem to be species-specific and are not found even with
our closest evolutionary relatives, the non-human primates. For
example, Tomasello (2008: 38) cites studies on ‘linguistic apes’, i.e.
human-raised apes which could communicate through signs, in which
96–98% of all signed productions were imperatives, with the remaining
2–4% having no clear functional interpretation but with no clear cases
of helpful information or narratives. Similarly, the data cited in
Håkansson and Westander (2013) contain no clear cases of non-human
animal communication where helpful information is being offered
(other than (largely involuntary) alarm signals) or narratives are
being told.
Hockett (1960 [1982]: 6) already notes that his design-feature of ‘displace-
ment’ or ‘being able to talk about things that are remote in space or time
(or both) from where the talking goes on’ is found not only in human
communication but also in bee-dancing. The referential communication
of bee-dancing, however, arguably does not correspond to what Toma-
sello (2008) refers to as ‘inform[ing] others of things helpfully’ (and
certainly not to the ‘narrative mode’) but presumably to a prompt or
command to other members of the hive to go to the place being
described. With that, the last two functions in Tomasello’s (2008) analy-
sis appear to be unique to humans.
18 For the original references, see the respective examples in Van Valin
(2005: 10).
19 ‘[. . .] in English, illocutionary force is indicated by the position of the
tense marker in the main clause: interrogative by core-initial tense,
declarative by core-internal tense, and imperative by no tense.’ (Van
Valin, 2005: 10, fn. 2).
20 See note 19.

21 Although the notion of finiteness is closely related to operators in RRG,

the two are not entirely identical, as, for example, RRG does not view
markers for honorificity (politeness, deferentiality) as operators,
although these are generally viewed as markers of finiteness in lan-
guages where these are found, e.g. Korean or in many South Asian
languages. For further discussion of finiteness from a typological per-
spective, see Maas (2004), Bisang (2001) and the studies in Nikolaeva
(2007).
22 On the use of this notation, see Van Valin (2005: 52).

3
A Conceptually Oriented
Approach to Semantic
Composition in RRG
Ricardo Mairal Usón and Pamela Faber
3.1 Introduction
This chapter discusses semantic representation in Role and Reference Gram-

mar (RRG) and how it can be enriched with a more conceptual or ontologic-
ally oriented approach.* An ontologically oriented model of semantic
representation would provide a framework for the formalization of mean-
ing construction. In fact, Van Valin and Mairal Usón (2014) argue that if the
RRG linking system is given a more conceptual orientation, this can sub-
stantially enhance its semantic representations. Frameworks such as Wierz-
bicka’s (1996) Natural Semantic Metalanguage and Pustejovsky’s (1995)
Generative Lexicon Theory have moved or are moving in this direction.
A case in point is the Brandeis Semantic Ontology (BSO) (Havasi et al. 2007;
Pustejovsky et al. 2006), a large lexicon ontology and lexical database. The
BSO depends on the Generative Lexicon (GL), a theory of linguistic semantics
that focuses on the distributed nature of compositionality in natural lan-
guage. This is an issue that RRG will also have to address.
Previous RRG research on the linking algorithm has primarily targeted the
syntactic level (layered structure of the clause) and paid somewhat less
attention to the semantic level (logical structure). However, semantic
decomposition is an important issue in RRG because its goal is to draw
empirically valid generalizations for as wide a range of languages as pos-
sible. A linguistic theory that aspires to universal applicability should base
meaning construction and semantic representation on a set of basic con-
cepts or near-primitives, which can be found in a wide range of languages.
Although there have been various proposals for a finer-grained representa-
tion of states and activities (Van Valin and Wilkins 1993; Van Valin and
* This research was funded by the Spanish Ministry of Economy and Competitiveness: grants FFI2011–29798-C02–01 and
FFI2014–53788-C3–1-P.

A Conceptual Approach to Semantic Composition 219
LaPolla 1997) and of active accomplishments (Van Valin, 2005, 2018; cf.
Bentley 2019), these issues still need to be explored in greater depth.
It goes without saying that the most basic function of language is to
convey meaning, and this involves considerably more than syntactic struc-
tures. At some level, an enriched lexical representation system would have
to account for the encyclopedic knowledge that speakers possess, which
enables them to choose the predicate that best represents the meaning they
wish to convey. For example, native speakers of English are intuitively aware
of the difference between related verbs such as those referring to loss of
possession (e.g. steal, purloin, pilfer, filch, swipe) or those referring to the
emission of a loud sound (e.g. scream, shriek, screech, bawl).
However, the ability of language users to differentiate between related
predicate meanings not only entails the consideration of single events (e.g.
drink) as compared to others in the same lexical domain (tipple, sip), but also
an awareness of the configuration of subevents designated by the same verb
within a larger ‘frame’ (Fillmore 1977, 1982). Frame Semantics (Fillmore
2006) studies how linguistic forms evoke frame knowledge, and how the
frames thus activated can be integrated into an understanding of the pas-
sages that contain these forms. Frames are crucial elements to consider in
the representation of lexical meaning.
For example, in certain cases, frames override constraints on the events
that a single predicate can encode. According to Croft (1991, 2012), the only
way in which two subevents can co-exist in the same verb is if they are
causally related (e.g. smash, which encodes the action of smashing as well as
its result). However, this is not necessarily true. Goldberg (2010: 44–49) states
that the verb double-cross designates an event of betrayal following a state or
event of understood cooperation. The betrayal is not caused by the state of
trust, nor does the betrayal cause the state of trust. Instead the state of trust
is part of the background frame that is presupposed in order for the profiled
or asserted act to count as double-crossing. This one subevent is profiled,
while the background frame presupposes one or more subevents without a
causal relation between them. Frame knowledge is thus essential to under-
standing the meaning of the verb.
A considerable part of a speaker’s language competence is the ability to
choose the right predicate and to select the most suitable arguments to fill
the slots activated by it in the communicative context. An effective lexical
representation system should be capable of accounting for differences in
lexical meaning within the context of a semantic domain and frame.
The first step in designing a representation for this purpose would
presumably involve the specification of a core set of basic concepts that
serves as a foundation for meaning representations. This is the only way to
obtain a formal representation with sufficient explanatory adequacy to
account for those aspects of meaning that transcend the code itself and
are part of the shared cultural knowledge and frames of a given speaker
community.

220 RICARDO MAIRAL USÓN AND PAMELA FABER
Indeed, RRG acknowledges the need for a more fine-grained system of

semantic representation. In fact, Van Valin and LaPolla (1997: 114) admit
that ‘there is as yet no adequate decompositional representation for the
primitive state and activity predicates which are the building blocks of the
system’. Despite the various proposals made to improve and enhance the
decompositional representation in RRG, none of them has as yet been
entirely satisfactory.
The rest of this chapter is organized as follows. Section 3.2 gives an
overview of semantic representation in RRG for all grammatical categories
and discusses semantic decomposition. Section 3.3 explores how logical
structures can be enhanced by conceptual information, more specifically
by the ontology in FunGramKB. Section 3.4 lists the most relevant conclu-
sions that can be derived from this chapter.
3.2 Semantic Representation in RRG
Since Foley and Van Valin (1984), the core component of RRG has been the
bidirectional linking algorithm that captures how syntactic and semantic
representations are mapped onto each other. This linking algorithm com-
prises a semantic level and a syntactic level, which model the communication
act between a speaker and an addressee. The semantic phase of the linking
algorithm begins in the lexicon, where a meaning representation is con-
structed from the information stored there. This representation is the input
for the syntactic phase, which involves the assignment of syntactic functions
and morphosyntactic properties. The syntactic phase, which proceeds in the
opposite direction, starts with an utterance. After the application of a syntac-
tic parser, the morphosyntactic properties of the input sentence are repre-
sented by means of the layered structure of the clause (LSC). This syntactic
representation is a combination of both universal and language-specific dis-
tinctions. The interpretation of the sentence stems from mapping rules that
link the syntactic representation to the semantic representation.
3.2.1 Semantic Representation of Predicates

The classification of predicates in RRG is an enriched version of Vendler’s
(1967) theory of Aktionsart. Van Valin (2005) augments the four basic classes
(state, activity, achievement and accomplishment) with one more class,
namely, semelfactives. He also distinguishes causatives as a parameter that
crosscuts these classes.1 Each of these classes is represented by a logical
structure (see Chapter 1). States and activities are primitives. In fact, they
are the definiens or genus in the representation of achievements, semelfac-
tives and accomplishments.
For example, a state predicate, such as see is represented by see′ (x, y). The
representation is composed of a primitive and two argument variables.

Activity predicates are expressed with do′,2 which is a characteristic of this

class of predicate, along with a primitive designating the activity carried
out. Activity predicates can have one argument (do′ (x, [hiccup′ (x)])), or two
arguments (do′ (x, [write′ (x, y)])). As previously mentioned, achievements,
semelfactives, accomplishments and active accomplishments are based on
states and activities in combination with operators.
More specifically, an achievement comprises a state or activity predicate
combined with INGR,3 indicating the sudden onset of the eventuality.
For instance, the vase cracked is represented as INGR cracked′ (vase). Semel-
factive predicates have a SEML operator and can be either activities or states.
The activity designated by these predicates is punctual but does not cause a
change of state or lead to a result. Thus, the janitor hiccupped would be SEML
do′ (janitor, [hiccup′ (janitor)])). A semelfactive predicate can also be a state,
for example flicker (SEML shine′ (x, y)).
Accomplishments are represented by means of BECOME, which signals
a change of state that reflects some sort of internal transition or trans-
formation, for example ossify: BECOME hardened′ (x). In Van Valin (2005:
44, 2018), BECOME is broken down into PROC, the operator which indi-
cates the duration of the change, and INGR, the operator which indicates
the change of state itself. Active accomplishments are a complex struc-
ture, composed of an activity and an achievement, which reflects the
telic nature acquired by the predicate in conjunction with a noun phrase
or an adpositional phrase. For example, the witch brewed a potion is repre-
sented as follows: do′ (witch [brew′ (witch, potion)]) & INGR exist′
(potion). This formulation also includes the ampersand (&), which means
‘and then’. Finally, each of these predicate classes also has a causative
version.4
Although verbs are extremely important, language competence also
involves other categories. The RRG treatment of adjectives, adverbs and
nouns is explained in Van Valin (2005: 48–51) and Van Valin and LaPolla
(1997: 154–171). When adjectives function as predicates, they are treated as
states in RRG logical structures (Van Valin 2005: 48).
(1) The chandelier is broken
broken′ (chandelier).
Adjectives can also occur as part of attributive predications. Note that be′ is
only found in the representation of attributive predications, not in the
representation of the resulting state; moreover, be′ is not the representation
of the copula, but the operator of a state, comparable to the operator do′.
The operator be′ thus figures in these representations as the principal
characteristic of attributive logical structures:
(2) a. Laura is wealthy.
be′ (Laura, [wealthy′])
b. John is afraid of birds.
feel′ (John, [afraid′ (birds)])

In regard to adverbs, Van Valin (2005: 49) explains that they are repre-
sented as one-place predicates and can modify different parts of a logical
structure. For example, the argument of a time adverb can be the entire
logical structure. In (3), the entire event of Lucas walking and then beginning
to be at the zoo takes place within the time frame designated by tomorrow.
(3) Tomorrow, Lucas will walk to the zoo.

tomorrow′ (do′ (Lucas, [walk′ (Lucas) ]) & INGR be-at′ (zoo, Lucas))
Generally speaking, manner adverbs modify activity predicates, whereas

pace adverbs can modify any type of durational predicate. In (4), the activity
of stealing, performed by Bugsy, is both an activity and durational predicate.
As a result, it can be carried out in a certain way (furtively) and at a certain
speed (fast):
(4) Bugsy furtively stole the jewels fast.

furtive′ (do′ (Bugsy, [steal′, (Bugsy, jewels))]) CAUSE [fast′ (BECOME stolen′
(jewels))]
Aspectual adverbs modify basic state or activity predicates. In (5), there is an

evident change of state since the water that was in the teapot is no longer in
a liquid state because it has been transformed into vapour.
(5) The water in the teapot totally evaporated.

BECOME (total′ (evaporated′ (water)))
In (6), a sustained activity is performed by the juror, who does not stop
moving throughout the time period of the stated event (trial).
(6) The juror fidgeted continuously.

continuous′ (do′ (juror, [fidget′ (juror)]))
Finally, Van Valin (2005) describes how predicative prepositions are repre-
sented and underlines the fact that they are also state predicates. As shown
in (7), the logical structure be-in′ is the highest predicate since the grassland
is the place where the kangaroo is jumping. As such, it has two arguments,
namely, the grassland and the logical structure for jump.
(7) The kangaroo jumped in the grassland.

be-in′ (grassland, [do′ (kangaroo, [jump′ (kangaroo)])])
This contrasts with the logical structure of the active accomplishment, the
kangaroo jumped to the grassland, in which the predicative preposition (PP)
indicates the final location of the kangaroo and not the location of the
jumping event.
3.2.2 Semantic Representation of Nouns

Nouns as a grammatical and semantic category have never been a focus in
RRG since their syntax is less complex. Up until now, RRG has proposed a
semantic representation based on the qualia analysis in Pustejovsky (1995).

Although, unlike verbs, nouns (e.g. cat and table) lack a logical structure, they
have semantic properties in the form of qualia roles, which parameterize
their meaning. According to Pustejovsky et al. (2006: 3), nouns can be
described in terms of the following:
• formal role: the basic type distinguishing the meaning of a word

• agentive role: the factors involved in the object’s origins or ‘coming
into being’
• telic role: the purpose or function of the object if there is one
• constitutive role: the relation between an object and its constituent parts.
Since the formal role refers to the type_of relation and the constitutive
role refers to the part_of relation, this is still another indicator of an
implicit typology of concepts underlying the semantic description of
nouns. Just as important is the fact that the agentive role identifies a set
of events linked to the object concept, whereas the telic role refers to an
event description, which is associated with that object as its function
(Pustejovsky et al. 2006: 333). According to Pustejovsky (2001: 5), qualia
provide functional tags for words, which are linked to the network of
concepts in Figure 3.1.
Within the context of this concept lattice or network, each entity is
described in terms of its qualia structure. For example, Figure 3.2 shows
the representation of the qualia structure of violin, which is classified as a
functional entity or artefact that has been created for a purpose. The
formal role (F) indicates that violin is a type of musical instrument. Its
agentive role (A) shows that a violin is the result of a building process. The
telic role (T) reflects that the purpose of a violin is to produce music;
and the constitutive role (C) indicates the parts of the violin, such as
the strings.
Entity Event Quality
Natural Functional Complex Natural Functional Complex Natural Functional Complex
Physical Abstract Direct Purpose
coffee knife book die eat read red heavy dangerous good rising frightened
Count Mass Info
Figure 3.1 Tripartite concept lattice (Pustejovsky 2001)

Figure 3.2 Qualia structure of violin (Pustejovsky and Batiukova 2019: 162)
The variables w, x, y and z refer to the different entities involved in events

typically associated with a violin. The violin itself is assigned the variable x
and is given a basic type in F; the agent building the violin is y (‘y builds x’),
as expressed in A. The person using the violin in T to produce the musical
sound on the violin is z (‘z produces sound on x’); and in C, the strings are
identified as w (‘w of x’) (Pustejovsky and Batiukova 2019: 162).
This overview of the RRG treatment of lexical units shows that words
belonging to all grammatical categories would benefit from a system of
semantic decomposition, which would make their meaning explicit and
highlight the relations between words in the same category as well as with
those in different categories.
For example, such a system would show that predicates describing
emotional states could be further decomposed in terms of feel, or those
predicates that express possession could be decomposed in terms of have.
Artefact entities such as table, gun and bread would be assigned to conceptual
categories of furniture, weapon and food, respectively. Humans would
also have different types of social roles (aunt, neighbour, friend, etc.) and
professional roles (lawyer, zookeeper, hairdresser, etc.).
Despite the fact that in RRG there is still no clearly defined system that
indicates which words should be treated as primitives and which should be
defined in terms of other more basic lexical units, this could be achieved by
specifying the meaning representations of lexemes and linking them to an
underlying ontological system.
3.2.3 Semantic Decomposition in RRG

Although the use of English words to designate potential primitives is a
notational convenience (Van Valin 2005: 46), the criteria used to select these
primitives are somewhat less clear. Even though lexical primitives may have
conceptual correlates, they still lack an ontology. There is also the need for
research that will provide evidence of the scope of semantic primitives or at
least give some indication of their potential cross-linguistic validity. This
would doubtlessly contribute to clarifying the lexical–conceptual interface
and how it works.
In the RRG lexical representation system, another issue is why certain
predicates undergo semantic decomposition whereas others do not. For
example, in the lexical representations for love (8) and snore (9), the same

predicates (love′ and snore′) are used in their own definitions. Nonetheless,
love, which is a verb of feeling, could be defined in terms of feel, whereas snore
is a type of sound emission.
(8) love love′ (x, y)
(9) snore do′ (x, [snore′ (x)])
The same also occurs in (10) and (11) except for the fact that the BECOME
operator introduces the resulting state.5
(10) soften BECOME soft′ (x)
(11) scatter [do′ (x, Ø)] CAUSE [BECOME scattered′ (y)]
These examples are in direct contrast to the representations in (12–14) in

which each predicate is semantically decomposed. More specifically, learn is
defined as to come to know, receive is to come to have, and show is to cause to
see. In this causative sense, x does something that causes y to come to see z.
(12) learn BECOME know′ (x, y)
(13) receive BECOME have′ (x, y)
(14) show [do′ (x, Ø)] CAUSE [BECOME see′ (y, z)]
According to Van Valin (1993) and Van Valin and Mairal Usón (2014), it is
necessary to improve the lexical representation system so as to obtain more
consistent semantic decompositions. However, part of the problem lies in
the fact that logical structures, as they were initially conceived, only focus
on capturing grammatically relevant aspects of word meaning and neglect
contextual meaning.
3.3 Towards a Conceptual Approach
Semantic decomposition of any kind is problematic because there is no

simple solution for the atomization of meaning. Nevertheless, semantic fea-
tures or attributes tend to pop up in different guises within a wide variety of
approaches, such as Wierzbicka’s Natural Semantic Metalanguage (NSM) (see
Goddard and Wierzbicka 2002), Mel’cuk’s Meaning Text Theory (e.g. Mel’cuk
1989, 2012), Jackendoff’s (1983, 1990, 1997) Conceptual Semantics, or Onto-
logical Semantics (see Nirenburg and Raskin 2004). This is also something that
Role and Reference Grammar has to deal with.
Even though semantic decomposition may at first appear to be a difficult
enterprise, the intuition persists that smaller meaning units must exist at
some level to encode conceptual content. This means organizing the lexicon
in lexical classes or domains. Each lexical domain has a basically hierarch-
ical organization with other types of relations linking each predicate to
others in a network structure. Factorization determines where the chain
of the decompositional system ends.

More specifically, conceptual information is encoded in a predicate defin-

ition, which to a certain extent is conceived as a frame with slots, having
certain selection restrictions and default values. In this respect, each verb in
the lexicon can be said to activate its own scenario, which determines its
semantic participants.
Designating a set of undefinables for a semantic metalanguage means the
adoption of a system of lexical decomposition which can possibly function
cross-linguistically. In this regard, each lexical domain has a set of functions
that act on the superordinate term to generate more specific hyponyms and
codify the most relevant subdomains. Ideally, the metalanguage established
through semantic decomposition could also be used for conceptual
representations.
A more ontologically driven representation can thus provide a solution to
the problems arising from the use in lexical representations of an unstruc-
tured inventory of semantic primes that are created as needed. Meaning
definitions within the lexicon can and should point to the position of a
concept within a network. This is in line with Pustejovsky (1995: 6) when
he writes
the meanings of words should somehow reflect the deeper conceptual

structures in the cognitive system, and the domain it operates in. This is
tantamount to stating that the semantics of natural language should be
the image of nonlinguistic conceptual organizing principles, whatever
their structure.
Such a conceptual configuration can be organized onomasiologically (in

meaning areas) rather than semasiologically (in alphabetical order). The
definitions in an ideal dictionary would thus reflect conceptual categories,
as codified in the genus or superordinate term of each definition. The
features or properties that distinguish concepts from others in the same
category would be linguistically represented by the clauses that modify the
genus. Definitions would be coherent not only on a microstructural level,
but also on a macrostructural one (Mairal Usón and Faber 2007).
According to Levy (2003), it is only through the study of the usage of
terms in a public language that we can have an independent way of
fixing the contents of people’s concepts. This coincides with Dummett’s
(1991) Priority Thesis, which takes language first and concepts second.
Since dictionaries are the codification of conceptual content in public
language, they are valid texts for the extraction of conceptual informa-
tion regarding meaning parameters, arguments, semantic roles, semantic
restrictions, etc.
It is well known that categories have a basically hierarchical organization,
given that hierarchies are central to cognition (Jackendoff 1997: 16). Within
such hierarchies, concepts are related both vertically and horizontally by
different types of conceptual relation. When this organization is applied to
the structure of a semantic or conceptual domain, the resulting structure is

an ontology (in the artificial intelligence sense rather than the philosophical
sense), defined by Gruber (1976) as an explicit specification of a conceptual-
ization. One crucial property of a conceptual system is that no concept can
be described without an account of its relationships to others (Lamb
1998: 147).
When such representations are anchored to a well-designed conceptual
ontology or network that relates word senses, the primitives in a conceptual
lexical structure are no longer regarded as predicates but rather as concep-
tual units taken directly from an ontology.6
The ontology that can be used to specify the concepts in conceptual logical
structures is FunGramKB, a multilingual online environment for the semi-
automatic construction of a multipurpose lexico-conceptual knowledge
base for natural language processing (NLP) systems (Periñan-Pascual and
Arcas-Tuñez 2007, 2010; Mairal Usón and Periñan-Pascual 2009).7 The struc-
ture of FunGramKB has two levels: a linguistic level and a conceptual level.
The three components of the linguistic level are the lexicon, grammaticon
and morphicon, which store the lexical, grammatical and morphological
information for each word. In a parallel way, the conceptual level is also
composed of three modules: the ontology (concept hierarchy), the cognicon
(scripts encoding procedural knowledge) and the onomasticon (proper
names of entities and events).
In the same way as other lexically based ontologies, the main premise of
FunGramKB is that the conceptual representations of objects, attributes and
events are mapped onto language in some significant way. The assumption
is that the structure of the lexicon is based on a core set of undefinables or
conceptual invariants and that these invariants can be extracted from
lexicographic resources that document our shared knowledge of the world.
Whereas the lexicon houses language-specific syntactic and morphological
information, the language-neutral ontology captures the meanings and
interrelationships of concepts that are not explicitly stated in the lexicon.
The first step in building the ontology was the specification of an inven-
tory of basic concepts that are assumed to be lexicalized in a wide range of
languages. The ontology acquisition methodology is based on the extraction
of type_of hierarchies from dictionary definitions (Martín-Mingorance 1984,
1990, 1995; Hirst 2009; Amsler 1980, 1981). Whereas the genus designates
the superordinate concept of the defined word, the differentiating features
are the properties that make the concept different from other members of
the same conceptual category. The meaning of a word is thus an access point
to a concept or conceptual structure of some kind. Pustejovsky (2001: 5) also
claims that his qualia are linked to a network of concepts (though this
relational structure is never specified). Consequently, all problems are trans-
ferred to the ontology since each word sense is represented as a pointer to
some concept or category within the ontology.
The FunGramKB ontology has the following levels. At the highest level
are meta-concepts (e.g. #EVENT). At the next level are basic concepts

(e.g. þHUMAN_00), which are used to define other basic concepts and
terminal concepts. At still another level are terminal concepts (e.g.
$METEORITE_00), which are more specific and not used in the meaning
postulates of other concepts. The final level of subconcepts (e.g.
PASTEURIZE), has the same thematic frame as a terminal concept (i.e.
$STERILIZE_00), but specifies one of its arguments to a greater degree. (In
the case of PASTEURIZE, the second argument is restricted to milk prod-
ucts.) The concepts at the meta-conceptual, basic, terminal, and subconcept
levels are preceded by the symbols, #, þ, $, and , respectively. Except for
the subconcept level, each also has a numerical code, (_00, _01, _002, etc.).
The methodology used to extract concepts consists of four phases: (1) con-
ceptualization, (2) hierarchization, (3) remodelling, and (4) refinement.8 The
source was the Longman Dictionary of Contemporary English (Procter 1978). All
basic and terminal concepts in FunGramKB are described in terms of seman-
tic properties represented in thematic frames and meaning postulates. Each
event is assigned to one thematic frame.
A thematic frame contains the number and type of participants (with
selection restrictions) involved in the event or quality. Although not all
participants are lexicalized, they are implicit to the understanding of the
concept. The second type of conceptual schema is a meaning postulate,
which is a set of one or more logically connected predications (e1, e2, etc.)
with the generic features of concepts. Only one thematic frame and one
meaning postulate are assigned to each concept.
Meaning postulates are encoded in COREL (Conceptual Representation
Language) (Periñán-Pascual and Mairal Usón 2010). This machine-readable
language defines all of the units. It can also be used to obtain conceptual
relations by applying inheritance and inference mechanisms to the mean-
ing postulates (Periñán-Pascual and Arcas-Túnez 2005). In each representa-
tion, the meaning postulates provide the basis for natural semantic
decomposition since each lexeme is linked to a concept.
3.3.1 Conceptual Logical Structures

Adopting this conceptual approach would significantly modify RRG in that
logical structures (LSs) would become conceptual logical structures (CLSs).
Although CLSs resemble LSs in that they maintain the Aktionsart distinc-
tions, the two differ because in CLSs, predicates become concepts. This type
of ontologically based representation has the format in Table 3.1.
As can be observed, in Table 1, the lexical primitives are replaced by
concepts (<C>), which provide a more granular semantic decomposition
and represent any conceptual unit, whether basic or terminal. One of the
advantages of this approach is that each lexical unit is linked to a concep-
tual unit and is enriched by the inheritance of the conceptual properties (i.e.
thematic frame and meaning postulate) of the related conceptual unit. As
widely discussed in Van Valin and Mairal Usón (2014), CLSs arise from the

Table 3.1 Inventory of conceptual logical structures
Predicate class Conceptual logical structure
State <C> (x) or (x, y)

Activity do′ (x, [<C> (x) or (x, y)])
Achievement INGR <C> (x) or (x, y), or
INGR do′ (x, [<C> (x) o (x, y)])
Semelfactive SEML <C> (x) or (x, y)
SEML do′ (x, [<C> (x) or (x, y)])
Accomplishment BECOME <C> (x) or (x, y), or
BECOME do′ (x, [<C> (x) or (x, y)])
Active accomplishment do′ (x, [<C> (x, (y))]) &
BECOME <C> (z, x) o (y)
Causatives α CAUSE β, where α and β are logical structures of any type
interaction of information in the ontology and the lexicon. In order to

ascertain how the ontology and the lexicon actually interact and arrive at
what has come to be known as the lexical–conceptual interface, it is neces-
sary to review what type of information is present in both the ontology and
the lexicon.
The choice of English as the conceptual language for lexical representa-
tion is not a problem since FunGramKB is based on the premise that at a
certain level, all languages contain a set of core concepts that lie at the heart
of communication. Since the ontology contains non-language-specific infor-
mation, the language used to encode concepts is irrelevant (in fact, numbers
could also have been used). Each concept derives its meaning from its
position in the ontology as well as from its relations with other concepts.
In other words, a concept is defined by its conceptual route. This is exempli-
fied in the representation of $STERILIZE _00 in Figure 3.3.
The ontology in Figure 3.3 represents $STERILIZE_00 as a material event
that is a type of transformation process, represented by the meta-concept
#TRANSFORMATION. At the more subordinate levels are the basic concepts,
þCHANGE_00 and þCLEAN_01. As shown, þCHANGE_00 is a type of trans-
formation, whereas þCLEAN_01 is a type of change in that an entity that is
dirty is changed, so that the dirt is removed and it becomes clean.
In Figure 3.3, $STERILIZE_00 is a daughter concept of þCLEAN_01. The
cleaning process in $STERILIZE_00 involves killing bacteria (or something
perceived to have bacteria). As such, $STERILIZE_00 is a terminal concept
with various predicates bound to it, namely, fumigate, sanitize, disinfect,
pasteurize, decontaminate and sterilize. Predicate selection is constrained by
the context. World knowledge (not syntax) pertaining to the entity with
bacteria, the sterilization instrument or method used, and the end result
pursued guide the speaker’s choice of predicate. Although the meaning
distinctions encoded in the predicates do not have an impact on syntactic
form, this type of semantic–pragmatic information should be included in
the RRG lexicon since it is a major component of lexical competence.

Figure 3.3 FunGramKB Ontological Hierarchy for $STERILIZE_00
Moreover, the fact that lexical entries are linked to conceptual units and
thus have different conceptual routes is a way of resolving polysemy. CLSs
facilitate the selection of the most suitable concept in the case of predicates
with more than one meaning. In this sense, each meaning of a predicate is
linked to a different concept. The result is a robust enumerative lexicon
with fine-grained meaning postulates, capable of generating a complex
conceptual network. It is thus no longer necessary to condense all the
meanings of a lexical unit into a single infra-specified representation as
proposed by Generative Lexicon Theory.
In CLSs, differences in meaning are not marked by the syntactic behaviour
of a lexeme but rather are specified in its meaning. Accordingly, nuances of
meaning resulting from differences in syntactic constructions do not affect
the conceptual representation or the selection of the conceptual unit. This
method of dealing with polysemy is exemplified by fix. For instance, one of
the various meanings of fix is to repair something that is broken or not
working properly (15).
(15) The plumber fixed the leaky tap.
Another meaning of fix is to prepare a meal or drinks.

(16) The chef fixed a dish of stewed artichokes.
FunGramKB links each of these senses to a different concept: þREPAIR_00

and þCOOK_00, respectively. Consequently, depending on the sense, fix has
one of the following formats:
(17) a. do′ (x, [þREPAIR_00 (x, y)])

b. do′ (x, [þCOOK_00 (x, y) & become <þEXIST_00> (y)]

The selection of (17a) or (17b) depends on the basic conceptual event since
cooking is a creation event whereas repairing is a change event. Accordingly,
þREPAIR_00 and þCOOK_00 differ in their conceptual routes and positions
in the ontology. In the case of þCOOK_00, the conceptual path is
#MATERIAL >> þDO_00 >> þCREATE_00 >> þCOOK. Its meaning postu-
late is the following:
(18) þ(e1: þCREATE_00 (x1) Theme (x2) Referent (f1: þHUMAN_00) Beneficiary
(f2) Instrument (f3: þFOOD_00)Means (f4: (e2: þHEAT_00 (x1) Theme (x3: f3)
Referent))Manner)
In contrast, the conceptual path of þREPAIR_00 is: #MATERIAL >>

#TRANSFORMATION >> þCHANGE_00 >> þREPAIR_00. However, the
change resulting from a þREPAIR_00 event is of a different type since it
involves changing an entity that was damaged in the past so that it can
function again.
(19) þ(e1: þCHANGE_00 (x1)Theme (x2)Referent (f1: (e2: past þDAMAGE_00

(x3)Theme (x2)Referent))Reason (f2: (e3: pos þOPERATE_00 (x4)Theme
(x2)Referent))Result)
The existence of a previously specified ontology provides a more specific

conceptual context that allows users to access concepts from a structured
inventory. There is thus no need to create ad hoc primitives on demand
since it is the ontology that determines the concept unit in the
representation.
3.3.2 Case Studies: pasteurize, obsolete and invoice

This section provides examples of the CLSs of verbs, adjectives and nouns.9
As previously mentioned, each lexical unit in the lexicon is linked to a
conceptual unit in the ontology.
For example, PASTEURIZE is a subconcept linked to the terminal concept,
$STERLILIZE_00 and as such, it is marked with a dash (). Its conceptual route
is #EVENT > #MATERIAL > #TRANSFORMATION > þCHANGE_00 >
þCLEAN_01 > $STERILIZE_00. It does not appear in the basic hierarchical
structure of the ontology since it shares the basic thematic frame and
meaning of $STERILIZE_00. The difference lies in the fact that in
PASTEURIZE, there is a specification of the second argument (milk). Words
in different languages can thus be linked to PASTEURIZE, such as pasteurize
(English), pasteurizar (Spanish), pastörize etmek (Turkish), 巴氏滅菌法
(Chinese). Example (20) shows a partial representation of the CLS for the
predicate pasteurize.
(20) pasteurize: do′ (x [PASTEURIZE (x, y)])
In (20), pasteurize is represented as an activity predicate that is the lexical

designation of the conceptual unit PASTEURIZE. However, this predicate
can be decomposed further. Figure 3.4 shows the conceptual entry for

Figure 3.4 FunGramKB conceptual entry for PASTEURIZE
PASTEURIZE, which consists of a thematic frame and a meaning postulate.

Thematic frames and meaning postulates are two complementary types of
conceptual schema used to encode the meaning of the concepts in
the ontology.
The thematic frame for PASTEURIZE is inherited from the terminal
concept $STERILIZE_00, which involves #TRANSFORMATION. In the
thematic frame of #TRANSFORMATION concepts, there are two participants:
(i) a Theme Entity that transforms another entity and (ii) a Referent Entity
that is transformed by another entity. However, the selection restrictions for
PASTEURIZE specify that the theme is human and the referent is restricted
to milk.
In Figure 3.4, the meaning postulate of pasteurize is composed of two
events, the second of which is the purpose of the other. In the second event
(e2), the theme (x1) kills a referent (x3), in this case, bacteria. Since
PASTEURIZE is a type of $STERILIZE_00, which entails a cleaning event
as well as a killing event, there is a semantic decomposition of this
predicate into more basic units. In fact, it is possible to reach a semantic
representation in COREL composed of semantic primitives by applying
stepwise conceptual decomposition (Periñán-Pascual and Arcas-Túnez
2005), where the defining term in a meaning postulate can become the
defined term in another. This process is based upon Dik’s (1978) stepwise
lexical decomposition as applied in Faber and Mairal Usón (1999). However,
instead of lexical items, the results of this process are regarded as
conceptual units.
This type of conceptual representation can also be used to represent
qualities designated by adjectives such as obsolete. The ontological hierarchy
in Figure 3.5 shows that obsolete is linked to the meta-concept of #QUALITY,
and that it exists as a social perception #SOCIAL. An entity can only be

Figure 3.5 FunGramKB Ontological Hierarchy for concepts linked to $OBSOLETE_00
Figure 3.6 FunGramKB conceptual entry for $OBSOLETE_00
perceived as þUSELESS_00 (!obsolete)) within a social context in which

something better has been created to replace it.
In the same way as with PASTEURIZE, words from different languages
can also be linked to $OBSOLETE_00 (e.g. Engl. obsolete, Sp. obsoleto, Fin.
vanhentunut, Hung. elavult). The CLS for these predicates has the following
format:
(21) $OBSOLETE_00 (x)
This CLS for obsolete consists of the conceptual unit $OBSOLETE_00, which,
in turn, is composed of a thematic frame and a meaning postulate
(Figure 3.6). Since it is a state, the thematic frame only has one participant
(x1), which is a previously useful entity that is now socially perceived
as useless.
Obsolete is a predicate that can be applied to a wide variety of objects, such
as typewriter. As is well known, a typewriter is a machine with keys that are

pressed to print letters of the alphabet onto paper. Since the 1990s, personal
computers more efficiently perform the functions of typewriters. Thus,
understanding why typewriters have become obsolete entails elements of
world knowledge, such as an awareness of past and present time frames as
well as of the affordances of typewriters in comparison to those of more
modern machines that can be used for the same purpose. The meaning
postulate in Figure 3.6 describes the three events that occur when an entity,
such as a typewriter, is described as obsolete:
(22) (e1: þBE_01 (typewriter)Theme (x2: $OBSOLETE_00) Attribute)

The typewriter is obsolete.
(23) (e2: þBE_01 (typewriter)Theme (x2: þUSELESS_00) Attribute)

The typewriter is useless (in the present).
(24) (e3: þBE_01 (typewriter)Theme (x2: þUSEFUL_00) Attribute)

The typewriter was useful (in the past).
As reflected in (22–24), $OBSOLETE_00 is a more specific type of

þUSELESS_00 since it presupposes a time sequence or period during which
the (once useful) entity has become useless. The representation in (25)
portrays þUSELESS_00 as the negation of þUSEFUL_00.
(25) *(e1: þBE_01 (x1)Theme (x2: þUSELESS_00)Attribute)

þ(e2: n þBE_01 (x1)Theme (x3: þUSEFUL_00)Attribute)
The conceptual entry for þUSEFUL_00 in (26) states that something is

useful when it helps a human to do something.
(26) þ(e1: þBE_01 (x1)Theme (x2: þUSEFUL_00)Attribute)

þ(e2: pos þHELP_00 (x1)Theme (x3: (e3: þDO_00 (x4: þHUMAN_00)Theme
(x5)Referent))Referent)
All of this information is inherited by $OBSOLETE_00. When specifically

applied to a typewriter, what is also activated is the list of affordances
provided, which other more modern machines now perform
more efficiently.
Finally, nouns, such as invoice, are also represented in terms of a
conceptual logical structure. From an ontological perspective,
þINVOICE_00 is a basic concept (see Figure 3.7). It is a member of the
meta-conceptual category of #PHYSICAL_OBJECT and more specifically,
that of #SELF_CONNECTED_OBJECT. Its conceptual path is
þINFORMATION_OBJECT_00! þWRITING_00! þDOCUMENT_00. An
invoice is a type of information object, which is informative because of
the writing in/on it, and which has the form of a document.
Again, lexemes in different languages, such as invoice (English), factura
(Spanish), faktura (Norwegian), ‫( ﻓﺎﺗﻮﺭﺓ‬Arabic), can be linked to this
conceptual entry, which has the following CLS:
(27) þINVOICE_00 (x)

Figure 3.7 FunGramKB Ontological Hierarchy for þINVOICE_00
Figure 3.8 FunGramKB conceptual entry for þINVOICE_00
As a noun, invoice does not have a thematic frame. However, this does not
mean that it lacks a relational structure. As a commercial document that
itemizes a transaction between a buyer and a seller, invoice indirectly refers
to a complex knowledge configuration, which speakers have to access in
order to understand the term. This information should be included in the
meaning postulate of the conceptual entry. Accordingly, the conceptual
entry for þINVOICE_00 in Figure 3.8 is composed of a meaning postulate
with two main events, which state what type of entity an invoice is (e1) and
the context in which it is produced (e2).

This second event contains three subevents (e3, e4, e5). Examples (28–31)
show each event and subevent within the corresponding frame, which
encodes the purpose of an invoice (f1) and the possible reasons for issuing
it (f2, f3).
(28) (e1: þBE_00 (x1: þINVOICE_00)Theme (x2: þDOCUMENT_00) Referent)

[An invoice is a document.]
(29) (e2: þGIVE_00 (x3: þHUMAN_00)Agent (x1)Theme (x3) Origin (x4:

þHUMAN_00) Goal)
[A human (x3) gives an invoice to another human (x4)]
(30) (f1: (e3: þPAY_00 (x4) Agent (x5)Theme (x6) Origin (x3) Goal))Purpose
[The purpose of an invoice is for x4 to pay x3.]
(31) (f2: (e4: past þSELL_00 (x3)Agent (x7)Theme (x8)Origin (x4)Goal))Reason ^

(f3: (e5: past þWORK_01 (x3)Theme (f4: x4) Goal))Reason
[The reason for issuing an invoice is because x3 sold something to x4 in the
past or because x3 worked for x4 in the past.]
As reflected in (28–31), an invoice activates knowledge that is linked to a

given sociocultural context. This shows that language does not exist in a void
but is closely related to the culture and society in which it is used. Evidently,
invoices would have little or no meaning for a tribe in the Amazon rainforest.
In contrast, invoices are crucial in a sociocultural environment that contem-
plates commercial transactions involving the emission of documents that
notify a receiver that payment is due for goods or services of some type. When
this kind of encyclopedic knowledge is included in a conceptual entry, it
provides users with the information necessary to understand the concept.
As can be observed, within the context of an underlying ontology, a rela-
tively simple representation, such as þINVOICE_00 (x), do′ (x [ PASTEURIZE
(x, y)]), or $OBSOLETE_00 (x), allows users to access a rich variety of encyclo-
pedic information regarding objects, processes and attributes. This is possible
because of the thematic frames, meaning postulates and inference mechan-
isms stored in the knowledge base.
In sum, the only difference from current LSs is that CLSs are made of
conceptual units and not lexical words. CLSs maximize informativeness and
minimize redundancy without losing the simplicity and elegance of the
formalism in LSs. A lexical entry is now defined within the framework of a
CLS, which is able to connect to a huge knowledge base from where it is
feasible to retrieve information via a reasoning engine, which in turn uses
two major functions: inheritance and inference (see Periñán-Pascual and Arcas-
Túnez 2010). By using artificial intelligence techniques, an RRG lexical repre-
sentation in terms of a CLS can be successively augmented by information
coming from different modules of the knowledge base (i.e. the ontology, the
cognicon and the onomasticon).10 Finally, a CLS is further enriched by the
inclusion of the operators that provide a fully-fledged semantic representa-
tion of an input sentence (Van Valin and Mairal Usón 2014).

3.4 Conclusion
RRG acknowledges the need for a more detailed theory of semantic repre-
sentation in order to justify semantic decomposition as well as the primitive
states and activities upon which it is based. This requires the formulation of
a representation that is not limited to syntactically relevant factors. Such a
representation system would be able to specify which lexeme would serve as
the generic term in the definition of a concept and would allow users to
effectively retrieve the semantic properties of each predicate in the lexicon.
RRG requires an ontological semantic theory, which is the natural evolution
of previous efforts to improve lexical representations, such as lexical tem-
plates with internal and external variables, based on lexical functions
and qualia.
For this reason, this chapter has provided a detailed description of a
conceptually oriented representation system that accounts for the non-
propositional dimension of meaning, and which allows users to access
contextual or encyclopedic meaning. In addition, it argues that adopting
an ontological approach in the form of CLSs instead of standard LSs in RRG
has numerous advantages. For example, CLSs can be automatically trans-
duced to a COREL scheme which can be used to obtain inference operations
with a reasoning engine. Consequently, the addition of CLSs to the RRG
system substantially enriches its semantic representations.
References
Amsler, Robert A. 1980. The Structure of the Merriam-Webster Pocket Dictionary.

PhD dissertation, University of Texas at Austin.
Amsler, Robert A. 1981. A taxonomy for English nouns and verbs. In Proceed-
ings of the 19th Annual Meeting of the Association for Computational Linguistics,
133–138. Morristown, NJ: Association for Computational Linguistics.
quantized change. International RRG Conference, 19–21 August 2019. Univer-
sity at Buffalo (SUNY).
Butler, Christopher S. 2012. An ontological approach to the representational
lexicon in Functional Discourse Grammar. Language Sciences 34: 619–634.
Cortés, Francisco, Carlos González and Rocío Jiménez. 2012. Las clases léx-
icas. Revisión de la tipología de predicados verbales. In Ricardo Mairal
Usón, Lilián Guerrero, and Carlos González (eds.), El funcionalismo en la
teoría lingüística. La Gramática del Papel y la Referencia. Introducción, avances y
aplicaciones, 59–84. Madrid: Akal.
Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago:
University of Chicago Press.
Croft, William. 2012. Verbs: Aspects and Causal Structure. Oxford: Oxford Uni-
versity Press.

Dik, Simon. 1978. Stepwise Lexical Decomposition. Lisse: de Ridder.

Dummett, Michael. 1991. Frege and Other Philosophers. Oxford: Oxford Univer-
sity Press.
Faber, Pamela and Ricardo Mairal Usón. 1999. Constructing a Lexicon of English
Verbs. Berlin: Mouton de Gruyter.
Fillmore, Charles J. 1977. Scenes-and-frames semantics. In Antonio Zampolli
(ed.), Fundamental Studies in Computer Science, 55–88. Dordrecht: North
Holland.
Fillmore, Charles J. 1982. Frame Semantics. In Linguistic Society of Korea
(ed.), Linguistics in the Morning Calm, 111–137. Seoul, South Korea: Hanshin
Publishing.
Fillmore, Charles J. 2006. Frame Semantics. In Dirk Geeraerts (ed.), Cognitive
Linguistics: Basic Readings, 373–400. Berlin: Mouton de Gruyter.
Goddard, Charles and Anna Wierzbicka. 2002. Meaning and Universal Gram-
mar: Theory and Empirical Findings. Amsterdam: John Benjamins.
Goldberg, Adele. 2010. Verbs, constructions and semantic frames. In Malka
Rappaport Hovav, Edit Doron and Ivy Sichel (eds.), Lexical Semantics, Syntax,
and Event Structure, 39–58. Oxford: Oxford University Press.
Gruber, Jeffrey S. 1976. Lexical Structures in Syntax and Semantics. New York:
North-Holland.
Havasi, Catherine, James Pustejovsky and Anna Rumshisky. 2007. An evalu-
ation of the Brandeis semantic ontology. Proceedings of the Fourth Inter-
national Workshop on Generative Approaches to the Lexicon. Paris, France.
Hirst, Graeme. 2009. Ontology and the lexicon. In Steffen Staab and Rudi
Studer (eds.), Handbook on Ontologies, 269–292. Berlin: Springer.
Jackendoff, Ray. 1983. Semantics and Cognition. Cambridge, MA: MIT Press.
Jackendoff, Ray. 1990. Semantic Structures. Cambridge, MA: MIT Press.
Jackendoff, Ray. 1997. The Architecture of the Language Faculty. Cambridge MA:
MIT Press.
Lamb, Sydney M. 1998. Pathways of the Brain: The Neurocognitive Basis of Lan-
guage. Amsterdam: John Benjamins.
Levy, Jacob T. 2003. Language rights, literacy, and the modern state. In
William Kymlicka and Alan Patten (eds.), Language Rights and Political
Theory, 230–249. Oxford: Oxford University Press.
Mairal Usón, Ricardo and Pamela Faber. 2007. Lexical templates within a
functional cognitive theory of meaning. Annual Review of Cognitive Linguis-
tics 5: 137–172.
Mairal Usón, Ricardo and Carlos Periñán Pascual. 2009. The anatomy of the
lexicon component within the framework of a conceptual knowledge
base. Revista Española de Lingüística Aplicada 22: 217–244.
Martín Mingorance, Leocadio. 1984. Lexical fields and stepwise lexical
decomposition in a contrastive English-Spanish verb valency dictionary.
In Reinhardt Hartmann (ed.), LEX’eter ’83 Proceedings. Papers from the

International Conference on Lexicography at Exeter, 226–236. Tübingen: Max

Niemeyer.
Martín Mingorance, Leocadio. 1990. Functional Grammar and Lexematics. In
Jerzy Tomaszczyk and Barbara Lewandowska-Tomaszczyk (eds.), Meaning
and Lexicography, 227–253. Amsterdam: John Benjamins.
Martín Mingorance, Leocadio. 1995. Lexical logic and structural semantics:
methodological underpinnings in the structuring of a lexical database for
natural language processing. In Ulrich Hoinkes (ed.), Panorama der Lexika-
lischen Semantik, 461–474. Tübingen: Gunter Narr.
Mel’cuk, Igor. 1989. Semantic primitives from the viewpoint of the
Meaning–Text Linguistic Theory. Quaderni di Semantica 10(1): 65–102.
Mel’cuk, Igor. 2012. Semantics: From Meaning to Text. Amsterdam: John
Benjamins.
Nirenburg, Sergei and Victor Raskin. 2004. Ontological Semantics. Cambridge,
MA: MIT Press.
Periñán Pascual, Carlos and Francisco Arcas-Túnez. 2005. Microconceptual-
knowledge spreading in FunGramKB. In 9th IASTED International Conference
on Artificial Intelligence and Soft Computing, 239–244. Anaheim-Calgary-
Zurich: ACTA.
Periñán Pascual, Carlos and Francisco Arcas-Túnez. 2007. Cognitive modules
of an NLP knowledge base for language understanding. Procesamiento del
Lenguaje Natural 39: 197–204.
Periñán Pascual, Carlos and Francisco Arcas-Túnez. 2010. The architecture of
FunGramKB. In Proceedings of the 7th International Conference on Language
Resources and Evaluation, pp. 2667–2674. Marrakech, Morocco: LREC.
Periñán-Pascual, Carlos and Ricardo Mairal Usón. 2010. La gramática de
COREL: un lenguaje de representación conceptual. Onomazein 21: 11–45.
Periñán-Pascual, Carlos and Ricardo Mairal Usón. 2011. The Coherent Meth-
odology in FunGramKB. Onomazein 24: 13–33.
Periñán-Pascual, Carlos and Ricardo Mairal Usón. 2012. La dimensión com-
putacional de la RRG: la estructura lógica conceptual y su aplicación en el
procesamiento del lenguaje natural. In Carlos González Vergara, Lilián
Guerrero Valenzuela and Ricardo Mairal Usón (eds.), El funcionalismo en la
teoría lingüística: la Gramática del Papel y la Referencia: Introducción, avances y
aplicaciones, 333–348. Madrid: Akal.
Procter, Paul. 1978. Longman Dictionary of Contemporary English. Harlow, UK:
Longman.
Pustejovsky, James. 2001. Type construction and the logic of concepts. In
Pierrette Bouillon and Federica Busa (eds.), The Language of Word Meaning,
91–123. Cambridge: Cambridge University Press.
Pustejovsky, James and Olga Batiukova. 2019. The Lexicon. Cambridge: Cam-
bridge University Press.
Pustejovsky, James, Catherine Havasi, Jessica Littman, Anna Rumshisky and
Marc Verhagen. 2006. Towards a generative lexical resource: The Brandeis

semantic ontology. Proceedings of the Fifth Language Resources and Evaluation

Conference, LREC 2006, 1702–1705. www.cs.brandeis.edu/arum/publica-
tions/lrec-bso.pdf.
Amsterdam Studies in Theory and History of Linguistic Science Series 4: 1–1.
Van Valin, Robert D., Jr. 2018. Some issues regarding (active) Accomplish-
ments. In Rolf Kailuweit, Lisann Künkel and Eva Staudinger (eds.), Applying
and Expanding Role and Reference Grammar, 71–93. Freiburg; Universität
Freiburg.
Van Valin, Robert D., Jr. and Ricardo Mairal Usón. 2014. Interfacing the
lexicon and an ontology in a linking algorithm. In María Ángeles Gómez,
Francisco Ruiz de Mendoza and Francisco Gonzálvez-García (eds.), Theory
and Practice in Functional-Cognitive Space, 205–228. Amsterdam: John
Benjamins.
Van Valin, Robert D., Jr. and David P. Wilkins. 1993. Predicting syntax from
semantics. In Robert D. Van Valin Jr., (ed.), Advances in Role and Reference
Grammar, 499–534. Amsterdam: John Benjamins.
Press.
Wierzbicka, Anna. 1996. Semantics: Primes and Universals. Oxford: Oxford
University Press.
Notes
1 See Cortés, González and Jiménez (2012) for an in-depth explanation of

the tests used to differentiate one predicate class from another.
2 RRG distinguishes between the primitive do′ and the operator DO in order
to represent the prototypical agency in verbs such as assassinate, perpetrate,
promise, etc., which require a human subject capable of volition: *The
cockroach assassinated the vice-president.
3 INGR stands for ‘ingressive’.
4 See Cortés, González and Jiménez (2012) for an explanation of how to
determine the Aktionsart of a predicate.
5 BECOME, which, as previously noted, is broken down into PROC INGR, is
used here for convenience.
6 Butler (2012) argues that a truly functional approach should provide a
rich account not only of lexical phenomena but also of conceptualization.
7 FunGramKB and related publications can be accessed at
www.fungramkb.com.

8 See Periñan-Pascual and Mairal Usón (2011) for an in-depth description of

the COHERENT methodology.
9 Predicative prepositions also have their corresponding conceptual
logical structures, which are obtained from the sub-ontology that stores
their corresponding conceptual units.
10 Periñán-Pascual and Mairal Usón (2012) have shown the applications of a
CLS in the area of machine translation and cross-linguistic information
retrieval.

4
Semantic Macroroles
Rolf Kailuweit
4.1 Introduction
Semantic macroroles are a major contribution of Role and Reference Gram-

mar (RRG) to linguistic theory. Foley and Van Valin (1984) introduced two
levels of semantic roles: an open list of specific thematic relations and two
generalized semantic roles, or macroroles, actor and undergoer. Specific
thematic relations were arranged in a graded Aktionsart-based continuum.
Further work in RRG (Van Valin 1993, 2005, 2010; Van Valin and LaPolla
1997) slightly modified the initial formalization. Dowty (1991) took up the
basic insights of Foley and Van Valin (1984), introducing the concept of
proto-roles (Kailuweit 2004). It should be acknowledged that Dowty (1991)
is currently the most prominent approach to generalized semantic roles1
(Rappaport Hovav and Levin 2015). However, approaches in typology (Has-
pelmath 2008; Bickel 2011) as well as in cognitive semantics (Jackendoff
2007) have built upon the RRG concept of macroroles to develop specific
semantic generalizations.
This chapter will deal with the theoretical foundations of macrorole
assignment in RRG. In the first part, the status of thematic relations in
RRG will be discussed in the context of a short historical review of Fillmore’s
(1968) deep cases and Gruber’s (1965 [1976]) theta-roles, as well as of the
discussion on generalized semantic roles in the 1980s and early 1990s. In the
second part, special attention will be paid to the role of Aktionsart, semantic
and syntactic transitivity, and causativity in macrorole assignment.
Although the focus of the chapter is on standard RRG, the third part of
the chapter discusses alternative approaches, concerning the number of
macroroles that should be postulated and the features that are pertinent
to macrorole assignment.

Semantic Macroroles 243
4.2 Thematic Relations
In RRG, semantic roles, called thematic relations, are understood as a

function of decomposed logical structures, following Gruber ([1965] 1976)
and Jackendoff (1972), in contrast with Fillmore’s (1968) lists of case roles.
Thematic relations are fundamental to making the concept of valency
fruitful for syntactic description. Tesnière ([1959] 1965), along with Hockett
(1958), one of the founders of the concept of valency, did not distinguish
clearly between syntactic and semantic functions in the definition of the up
to three actants a predicate could take. Hence, Tesnière ([1959] 1965) was not
able to account for the passive construction in a coherent way.2 As was
pointed out by Ruwet (1972), this is only possible if the levels of syntactic
functions and thematic relations are clearly separated. Thematic relations,
such as agent and patient, refer to the semantics of the predicate’s argu-
ments, which remain stable, while the syntactic realization of the argu-
ments varies between the active and the passive construction.
(1) Peter ate an apple.

[Peter ¼ agent, apple ¼ patient]
(2) The apple was eaten by Peter.

[Peter ¼ agent, apple ¼ patient]
Thematic relations are not just isolated representations of the individual

arguments that are taken by a predicate. They are sets of labels, one for each
argument, with each label specifying the relation of that argument in the
event in comparison to the other arguments. However, research in line with
the initial approaches of Gruber ([1965] 1976) and Fillmore (1968) failed to
determine the criteria to establish a coherent set of thematic relations.
Thematic relations such as agent, patient or theme, etc. were supposed
to be systematic generalizations across verb-specific roles such as runner
(for run′), killer and killed (for kill′), possessor and possession (for
have′), judger and judgement (for consider′), etc. Nonetheless, the
number of thematic relations varied considerably in the literature, ranging
from six (agent, theme, location, source, path, goal) in rigorous
localist approaches to a few dozen in other approaches (see Rauh 1988;
Dowty 1991; Fillmore 2003; Rappaport Hovav and Levin 1988; see Rappaport
Hovav and Levin 2015: 600 for an overview). Fillmore introduced several non-
localist roles – deep structure cases in his terminology – which have been
widely discussed in the literature: instrument, force, experiencer,
beneficiary, etc. However, while natural forces are participants of very
specific events, experiencers are generalizations at a more abstract level,
referring to arguments of verbs of emotion, perception and cognition. The
difficulties in elaborating a coherent inventory of thematic relations led
some authors to do away with thematic roles as categories of
semantic description.

244 ROLF KAILUWEIT
Rappaport Hovav and Levin (1988), Grimshaw (1990) and Tenny (1994)
advocated an aspectual approach. According to these authors, the aspectual
properties of a predicate alone determine the syntactic realization of the
arguments. Telicity was deemed to be a pertinent criterion: an argument
affected by a telic process has to be realized as an object at some level of
representation.3 However, Rappaport Hovav and Levin (2015: 606) pointed out
that change-of-state verbs, the most stable class in terms of their linking
properties, are aspectually heterogeneous, including telic or variably telic
verbs (degree achievements according to Dowty (1979), e.g. flatten versus
widen).4 Moreover, no aspectual criterion is available to predict the linking
of two-place state predicates such as those in (3) and (4).
(3) Peter abstained from alcohol and smoking.
(4) Mary believed in justice.
Hence, not only aspectual properties, but also non-aspectual properties such
as sensation, perception, volition and control, play an important role in argu-
ment linking. According to François (1997), aspectual as well as participant-
oriented properties (sensation, perception, control and causation5) are indis-
pensable for argument linking. Macroroles in RRG as instantiations of general-
ized semantic roles are a construct that allows such an integration.
4.3 Macroroles as Generalized Semantic Roles
Foley and Van Valin (1984) considered macroroles to be generalizations

about thematic relations. After developing logical structures (LS) as a for-
malism for decompositional representation of verb semantics,6 they based
macrorole assignment on both LS and thematic relations (ibid., p. 53). As Van
Valin (1999) pointed out, later versions of RRG denied thematic relations
theoretical status, arguing that thematic relations are redundant for lexical
representations.7 Thematic relation labels of the kind we find added to LS in
Table 4.1 are only maintained for convenience as mnemonics (Van Valin and
LaPolla 1997: 128).
In RRG the relevant generalizations for macrorole assignment do not
operate directly over thematic relations but over argument positions in
logical structures representing activities and states, as can be seen in
Table 4.1.8 Importantly, the criteria applied in macrorole assignment are
not entirely aspectual. This becomes evident if we compare the two leftmost
positions of the Actor-Undergoer Hierarchy (AUH) illustrated in Table 4.2.
Both positions imply an activity component (DO (x,. . .)) or (do′ (x,. . .)). The
argument of DO is a controlling human argument: for example, the first
argument of murder. In contrast, the first argument of kill need not have the
referential value þhuman. A natural force (e.g. hurricane) or an abstract
concept or state of affairs (e.g. disease, drinking so much) can be the first
argument of (do′ (x,. . .)).9

Table 4.1 Logical structures of states and activities and thematic relations (see Van
Valin and LaPolla 1997: 115)
I. STATE VERBS
A. Single argument
1. State or condition broken′ (x) x ¼ PATIENT
2. Existence exist′ (x) x ¼ ENTITY
B. Two arguments
1. Pure location be-LOC′ (x, y) x ¼ LOCATION , y ¼ THEME
2. Perception hear′ (x, y) x ¼ PERCEIVER , y ¼ STIMULUS
3. Cognition know′ (x, y) x ¼ COGNIZER , y ¼ CONTENT
4. Desire want′ (x, y) x ¼ WANTER , y ¼ DESIRE
5. Propositional att. consider′ (x, y) x ¼ JUDGER , y ¼ JUDGEMENT
6. Possession have′ (x, y) x ¼ POSSESSOR , y ¼ POSSESSED
7. Internal experience feel′ (x, y) x ¼ EXPERIENCER , y ¼ SENSATION
8. Emotion love′ (x, y) x ¼ EMOTER , y ¼ TARGET
9. Attrib./Identific. be′ (x, y) x ¼ ATTRIBUTANT , y ¼ ATTRIBUTE
II. ACTIVITY VERBS

A. Single argument
1. Unspecified action do′ (x, Ø) x ¼ EFFECTOR
2. Motion do′ (x, [walk′ (x)]) x ¼ MOVER
3. Static motion do′ (x, [spin′ (x)]) x ¼ ST - MOVER
4. Light emission do′ (x, [shine′ (x)]) x ¼ L - EMITTER
5. Sound emission do′ (x, [gurgle′ (x)]) x ¼ S - EMITTER
B. One or two arguments
1. Performance do′ (x, [sing′ (x, (y))] x ¼ PERFORMER , y ¼ PERFORMANCE
2. Consumption do′ (x, [eat′ (x, (y))]) x ¼ CONSUMER , y ¼ CONSUMED
3. Creation do′ (x, [write′ (x, (y))]) x ¼ CREATOR , y ¼ CREATION
4. Directed perception do′ (x, [see′ (x, (y))]) x ¼ OBSERVER , y ¼ STIMULUS
5. Use do′ (x, [use′ (x, y)]) x ¼ USER , y ¼ IMPLEMENT
The other three positions are related to a stative component in the LS

(pred′ (x, y) or pred′ (x)). RRG distinguishes between one-place and two-
place states. The argument of a one-place state is not only an entity that
exists, but also and most prototypically an entity undergoing a change of
state, being destroyed or killed, or coming into existence. Hence, the
aspectual properties of accomplishments and achievements, which
macrorole assignment does not directly refer to, are indirectly repre-
sented insofar as (pred′ (x)) prototypically formalizes the last component
of the LS of achievements and accomplishments.10 Note that not all one-
place states are represented in the lexicon as monovalent verbs. Most
typically, one-place states appear as part of the semantic representation
of causation. Causative verbs show a complex semantic representation
with both a cause and an effect component (αCAUSEβ). Causativity itself
is not represented in the AUH. As Van Valin (2005: 57) points out: ‘the
role assignments of the causative logical structure are those of the
constituent activity and other logical structures; no new roles are added.
The effect component (. . .CAUSEβ) prototypically consists of a resultative
one-place state. As an effect of a breaking event, for example, an entity
becomes broken.’

246 ROLF KAILUWEIT
Table 4.2 Actor-Undergoer Hierarchy (see Van Valin and LaPolla 1997: 127, 146;
Van Valin 2005: 61)
A CTOR U NDERGOER
argument 1. argument of do′ 1. argument of 2. argument of argument of

of DO (x,. . .) pred′ (x, y) pred′ (x, y) pred′ (x)
AGENT EFFECTOR LOCATION THEME PATIENT

MOVER PERCEIVER STIMULUS ENTITY
ST - MOVER COGNIZER CONTENT
L - EMITTOR WANTER DESIRE
S - EMITTOR JUDGER JUDGEMENT
PERFORMER POSSESSOR POSSESSED
CONSUMER EXPERIENCER SENSATION
CREATOR EMOTER TARGET
SPEAKER ATTRIBUTANT ATTRIBUTE
OBSERVER PERFORMANCE
USER CONSUMED
CREATION
LOCUS
IMPLEMENT
¼ increasing markedness of realization of argument as macrorole
Two-place states (pred′ (x, y)) formalize the semantics of bivalent stative
verbs of possession or emotion such as own or like. In line with localist
approaches (Gruber [1965] 1976; Jackendoff 1972), the x-argument of (pred′
x, y) could be thought of informally as a specific instance of a location. The
y-argument of (pred′ x, y) informally corresponds to specific instances of a
theme in terms of Gruber ([1965] 1976).
To sum up, the AUH represents a continuum of argument positions in
different subparts of lexical representations. The leftmost argument is the
prototypical candidate for actor, the rightmost argument the prototypical
candidate for undergoer.11 By default, the number of macroroles follows
from the LS. For lexical representations with no arguments, for example
rain′, no macrorole will be assigned. One-place intransitive activities, for
example do′ (x, [shine′ (x)]), will take an actor. One-place intransitive non-
activities, for example exist′ (x), will take an undergoer. Macroroles always
correspond to direct core arguments, although not every direct core argu-
ment bears a macrorole.12 By default, transitive predicates with two or three
arguments take two macroroles. The default macrorole assignment
principles are the following.

number of arguments in its LS:

b. Nature: for verbs which take one macrorole:

(Van Valin 2013: 78)
(6) a. The hunter [A(ctor)] killed the [do′ (x, Ø]CAUSE[BECOME dead′ (y)].13
bear [U(ndergoer)]
b. The child [A] saw her friends [U] do′ (x, [see′ (x, y)])
c. The girl [A] was running do′ (x, [run′ (x)])
d. The stars [A] were shining do′ (x, [shine′ (x)])
e. The ghost [U] did not exist exist′ (x)
However, there are important cases where the number of macroroles is

not entirely predictable from the LS. The LS have′ (x, y) corresponds to two-
place predicates of possession independently of whether these predicates are
syntactically transitive (the own-type, e.g. German besitzen) or intransitive
(the belong to-type, e.g. German gehören). The same holds for two-place predi-
cates of liking. The LS like′ (x, y) corresponds to transitive predicates (English
like, French aimer bien, German mögen, etc.) as well as intransitive predicates
(French plaire, German gefallen, etc.). In such cases, macrorole intransitivity
has to be marked in the lexicon.
(7) a. German gefallen like′ (x, y) [MR1]

b. German gehören have′ (x, y) [MR1]
For transitive verbs, the default macrorole assignment rules predict the
number of macroroles but do not stipulate which argument takes the actor
macrorole and which argument takes the undergoer macrorole. The default
choice follows from the consideration that ‘given the LS for a transitive verb,
the leftmost argument in it will be the actor and the rightmost will be the
undergoer’ (Van Valin 2013: 78).
It follows from the examples in Van Valin (2013: 78) that this also holds for
causatives with two LS components (cause and effect) linked together with a
CAUSE operator. Starting from the left with the αCAUSE component of the
LS, in example (8) the actor macrorole is assigned to the x-argument (dog).
Undergoer assignment will start from the right of the CAUSEβ component.
The rightmost accessible argument will be the y-argument (boy) that will be
selected for undergoer.
(8) The dog scared the boy. [do′ (x ¼ dog, Ø] CAUSE [feel′ (y ¼ boy, afraid′])]).
Note that if the undergoer were always the rightmost argument, predicates
represented by a causative structure with a two-place state in the CAUSEβ
component should always select the second argument of pred′ (x, y) as
undergoer.
(9) give [do′ (x, Ø)] CAUSE [BECOME have′ (y, z)]
(10) load [do′ (x, Ø)] CAUSE [BECOME be-on′ (y, z)]

248 ROLF KAILUWEIT
However, some three-place predicates of transfer (give), putting (load, spray),

or removal (drain, empty) show variable undergoer choice. Depending on the
language, the verb subclass, and, in the last analysis, the individual
lexeme,14 some predicates allow the first argument of the two-place state
in the CAUSEβ component (pred′ (x,. . .)) to take the undergoer macrorole
(e.g. Mary gave Peter a book. Max loaded the truck with hay).
In standard RRG, this pattern was considered to be a marked undergoer
choice. The German applicative construction provides a theory-independent
argument for the markedness of the pred′ (x,. . .)-as-undergoer construction.
While the morphological unmarked laden (‘load’) realizes pred′ (. . .,y)) as
undergoer, the pred′ (x,. . .)-as-undergoer construction requires additional
marking with the morpheme be-.
(11) a. Max lädt die Kisten auf den Laster

Max load.3sg the box.f.pl.acc on the truck
‘Max loads the boxes on the truck.’
b. Max belädt / *lädt den Laster mit Kisten
Max be.load3sg load3sg the truck.acc with box.f.pl.dat
‘Max loads the truck with boxes.’
Van Valin (2005: 126) acknowledged that from a strictly typological view-
point, the markedness hypothesis has to be abandoned due to the fact that
in primary object languages15 the pred′ (x,. . .)-argument seems to be the
unmarked choice for undergoer (see Guerrero and Van Valin 2004; Diedrich-
sen 2008). While most languages follow principle A in (12), some languages
follow principle B.
(12) Principles for Undergoer-choice (Van Valin 2005: 126)
1. Principle A: lowest-ranking argument in LS (default)
2. Principle B: second-highest-ranking argument in LS
Obviously, one could also apply these principles to different constructions in

one and the same language, in which case the pred′ (x,. . .)-as-undergoer
construction of English give or load would no longer be an instance of
marked undergoer choice but of principle B in a language that generally
follows principle A.
In Foley and Van Valin (1984: 29), macroroles were defined in purely
semantic terms, actor as ‘the argument which expresses the participant
which performs, effects, instigates or controls the situation denoted by the
predicate’, and undergoer as ‘the argument which expresses the participant
which [. . .] is affected by it in some way’. However, macrorole (in)transitivity
at the lexical level and variable undergoer choice show that macrorole
assignment is not only determined by position in LS but also by the semantic
properties which determine whether a core argument is realized as a direct
core argument in the morphosyntax.16 Indirect core arguments cannot be
undergoers. From the RRG point of view, it is because of the semantic
entailments that they get from verbs that core arguments are realized as
direct or indirect.

If undergoer selection is flexible, at least for some predicates, then one

might be tempted to assume that macrorole assignment parallels the basic
claim of the so-called (neo-)constructionist approach. In her syntactic
predicate-based approach Borer (1994, 2005) claims that semantic interpret-
ation derives mostly from the syntactic realization of arguments. There is
evidence that, at least from the hearer’s perspective, the syntactic configur-
ation entwined with the undergoer choice determines meaning (see Rappa-
port Hovav and Levin 2015: 615–616 for relevant discussion). However, in the
neo-constructionist perspective, the meaning is assigned by a syntactic
substructure, in RRG the meaning derives from the lexically determined
and in some cases flexible assignment of undergoer, that is, a semantic
category playing a role in syntax (Van Valin 2004: 74–78).
Let us consider again the case of the locative alternation. The semantics of
the two constructions with load are not equivalent. According to Anderson
(1971), there is always a holistic interpretation of the direct object (i.e. the
undergoer). All of the hay is loaded onto the truck when the pred′ (. . .,y)-
argument is chosen as undergoer and appears in the direct object position,
and the truck may or may not be full. When the pred′ (x,. . .)-argument is
chosen as undergoer and as direct object, the truck is completely filled,
without any implication as to whether all of the hay is loaded or not (cf. Van
Valin 2007). Kailuweit (2008: 350) described the semantic effects of marked
undergoer choice in the following way: if the x-argument of a transitive
predicate pred′ (x, y) is exceptionally chosen as undergoer, its interpretation
does not rely on its inherent semantic properties. It is the undergoer func-
tion that contributes certain prototypical semantic values to the argument,
that is, at least one of the following properties: ‘causally affected’, ‘change of
state’, or ‘incremental theme’. In psycholinguistic terms, one could argue
that the linking from syntax to semantics reveals that the ‘wrong’ argument
in the LS has been chosen for undergoer. This leads to cognitive dissonances
that can only be resolved by referring to the semantics prototypically associ-
ated with the undergoer. Hence, the pred′ (x,. . .)-argument is not just
conceived as a location, but as an argument primarily affected in the
event. The truck the hay is loaded onto is interpreted as being completely
filled, which is not the case when the pred′ (. . .,y)-argument, the hay, is
chosen as undergoer.
To sum up, RRG macroroles are neither mere generalizations of thematic
relations, nor is their assignment entirely motivated by aspectual (Aktion-
sart) differences. While the basic Aktionsart opposition between activities
and states plays a crucial role, other aspects are pertinent, too. Not only do
the proto-agent properties ‘volition’, ‘sensation’ and ‘control’17 help to dis-
tinguish a prototypical agent (argument of DO) from a mere effector (first
argument of do′ (x,. . .)), but they also implicitly motivate the order of the
arguments in two-place states. Proto-agent properties characterize a per-
ceiver, cognizer, wanter, judger, possessor, or experiencer argu-
ment and justify placing it at a higher level of activity (pred′ (x,. . .)) in

250 ROLF KAILUWEIT
comparison with a stimulus, content, desire, judgement, possessed,

or sensation argument (pred′ (. . .,y)).18 The processing rule, which pro-
cesses logical structures from left to right (or starting from the element that
comes first in time), is a syntactic device to cope with causal chains such as
(agent > instrument). Since RRG presupposes identical LSs for transitive
and intransitive predicates of possession or liking, macrorole transitivity is
affected by the semantic properties which result in the morphosyntactic
realization of core arguments as direct or indirect. Finally, to our mind,
variable undergoer choice adds an element to the projectionist RRG
approach that shows interesting parallels with the neo-constructionist
approach. The undergoer function, although semantic in nature, can over-
ride the semantics associated with the argument position of the LS of a
given entry in the lexicon. However, variable undergoer choice is not a
putative structural option for any three-place predicate, but a semantic
property determined by the lexicon.
4.4 Putative Challenges to Macrorole Assignment
In this section, we deal with some types of predicates that can be said to be
challenging for the RRG theory of macrorole assignment.
4.4.1 Causativity
When Foley and Van Valin (1984) first proposed the AUH, they presupposed –
in line with Dowty (1979) – a logical link between causative constructions
and accomplishments. In addition, they considered the argument of
αCAUSE to be an argument of DO. As the argument of αCAUSE is necessarily
the most active argument in the hierarchy, no problem arose with macro-
role assignment.
(13) Joan broke the glass [DO (Joan)] CAUSE [BECOME broken′ (glass)]
(Foley and Van Valin 1984: 52)
However, later research in RRG (Van Valin and LaPolla 1997: 97, 106–109;
Van Valin 2018) has shown that equating causative constructions with
accomplishments was erroneous. On the one hand, there are accomplish-
ments that are not causative, like the one in (14a), which contrasts with its
causative counterpart in (14b).
(14) a. The ice melted (accomplishment)

b. Hot water melted the ice (causative accomplishment)
On the other hand, non-causative states, activities and achievements can

also have corresponding causative constructions.
(15) a. The boy is afraid (state)

b. The ball bounced around the room (activity)
c. The balloon popped (achievement)

(16) a. Bill’s owning a gun frightened Martha (causative state)

b. The girl bounced the ball around the room (causative activity)
c. The cat popped the balloon (causative achievement)
The representation of causativity did not pose formal macrorole assign-

ment and linkage problems for Foley and Van Valin (1984). Perhaps for this
reason, the interaction of the AUH and causative constructions has not
been discussed in detail in later versions of RRG. Some putatively problem-
atic cases for macrorole assignment that arise from the isolated applica-
tion of the AUH for the individual predicates in causative constructions
can easily be ruled out if macrorole assignment begins with the actor and
simply occurs in the causal chain from left to right. These cases are
discussed below.
We will start with causative activities. Van Valin and LaPolla (1997: 107)
propose the LS [do′ (girl, Ø)] CAUSE [do′ (ball, [bounce′ (ball)])] for (16b). Both
components of this LS consist of activities. Hence, the two arguments (girl
and ball) occupy the same position in the hierarchy, but the processing rule
assures that the actor is assigned to girl. Being the rightmost argument, ball
in the CAUSEβ part takes the undergoer macrorole, although it is an argu-
ment of do′ (x,. . .). Obviously, this a coherent interpretation from the
semantic point of view, since the CAUSEβ component is the effect of the
causing event.
This is also true for activities embedded into causative chains (Van Valin
and LaPolla 1997: 121). The AUH explains why knife could eventually become
actor in an instrument-subject construction (cf. 17b).
(17) a. Tom is cutting the bread with a knife

[do′ (Tom, [use′ (Tom, knife)])] CAUSE [[do′ (knife, [cut′ (knife, bread)])]
CAUSE [BECOME cut′ (bread)]]
b. The knife cuts the bread
[do′ (knife, [cut′ (knife, bread)])]
A proposal to make the possible undergoer assignment to arguments of

do′ (x,. . .) more explicit would be to mark this position as accessible for
undergoer choice in the AUH (Kailuweit 2012: 110) (see Table 4.3).
In (16a), an example of a causative state, macrorole assignment does not
follow straightforwardly from the AUH. Van Valin and LaPolla (1997: 107)
Actor Undergoer
argument of 1. argument 1. argument 2. argument argument of

DO of do′ of pred′ of pred′ pred′ (x)
(x,…) (x, y) (x, y)
= increasing markedness of realization of argument as macrorole

Figure 4.1 Actor-Undergoer Hierarchy (modified)

252 ROLF KAILUWEIT
propose the LS [have′ (Bill, gun)] CAUSE [feel′ (Martha, [afraid′])] for (16a).
Both compounds of CAUSE consist of a two-place state predicate (have′(x ,y)
and feel′ (x, y)). For the β part the y-argument is a predicate [afraid′]. The two
x-arguments of these predicates (Bill and Martha) are both instances of the
same middle position of the AUH. If we follow the rule of processing from
left to right, Bill will be the actor. However, this is inaccurate from a
semantic point of view. It is not Bill, but his possession of a gun that is
responsible for causing Martha’s fear. Hence, the actor of frighten is not Bill,
but the whole αCAUSE compound. It may seem strange to think of a state of
affairs as an actor, but otherwise frighten would cease to be a macrorole
transitive predicate in (16a).
4.4.2 Three-Place Predicates

According to Van Valin (2007), almost all three-place predicates are causative
constructions. We have noted that these predicates give rise to problems of
variable undergoer choice. The overall LS of these predicates would be [do′
(x, Ø)] CAUSE [BECOME pred′ (y, z)]. The actor is represented in the αCAUSE
part and the CAUSEβ part hosts the remaining two arguments that are
accessible for undergoer assignment. As we have already seen, the condi-
tions of undergoer assignment vary across languages and even across indi-
vidual predicates. Nonetheless, there are a few (but highly frequent) three-
place predicates that do not seem to be causative. In the remainder of this
section, I will deal with talk (to somebody about something)19 as a candidate for
a possible non-causative three-place activity and with envy (somebody (for)
something) as a candidate for a possible a non-causative three-place state.
In Van Valin and LaPolla (1997: 116–118) we find a detailed description of
English verbs of saying, including verbs of talking. The authors propose a
general LS for non-causative verbs and an LS for causative verbs of telling.
(18) talk: [do′ (x, [express.(α).to.(β).in.language.(γ)′ (x, y)])]

tell: [do′ (x, [express.(α).to.(β).in.language.(γ)′ (x, y)])] CAUSE [BECOME
aware.of′ (y, z)], where y ¼ β, z ¼ α
Talk is listed as a verb that selects an α component and a β component (talk to

Pat about Sandy) or a γ component (talk Cajun). Since it is the absence of
causativity that distinguishes verbs of talking from verbs of telling, the
causative LS in (18) cannot describe the three-place use of talk. In a later
chapter concerned with coreference, Van Valin and LaPolla (1997: 401)
propose an LS for talk, when they discuss the coreference of the reflexive
in Bob talked to Susan about herself.
(19) [do′ (Bob, [express.(α).to.(β).in.language.(γ)′ (Bob, Susani)])] ᴧ [about′ (herselfi)]
The binding facts seem to be the reason why the β component appears as
the y-argument of the two-place state and the α component as the argument
of the preposition co-referring with the y-argument. In a footnote Van Valin

and LaPolla (1997: 668) remark that about herself is an argument PP and not
an argument-adjunct, because the PP is ‘the optional realization of the α
variable’. The fact that the preposition seems to be semantically empty and
commutable (talk about, talk of as well as German sprechen über, sprechen von)
corroborates this view. However, it is precisely for this reason that it seems
problematic to use an argument-adjunct construction to represent the third
argument of talk.20
Staudinger et al. (2008) proposed an alternative analysis. RRG allows for
several classes of two-place activities: do′ (x, [sing′ (x,(y))]); do′ (x, [see′ (x,(y))]);
do′ (x, [tap′ (x,(y))]), etc. (Van Valin and LaPolla 1997: 115).21 The second
argument represents an instance of pred′ (x, y) (ibid.: 127), but the predicate
embedded in activities is not a state. Hence, RRG could theoretically allow
for three-place activities with an argument in each of the three medium
positions of the AUH: do′ (x), pred′ (x, . . .), and pred′ (. . ., y). Staudinger et al.
(2008) thus suggested the following LS for French parler, the counterpart of
English talk.
(20) parler à qn de qc ¼ do′ (x, [talk′ (x, y, z)])
Note that according to the formalism used in RRG, the predicate talk′
(x,(y),(z)) is not a state, similarly to sing′ (x, (y)) in do′ (x, [sing′ (x, (y))]). As
indicated before, activities do not consist of states embedded under a do′
operator. Hence, in an RRG description of talk, one could associate the two
arguments y and z with the positions pred′ (x, . . .) and pred′ (. . ., y) of the
AUH, as RRG associates the y-argument of sing′ with the pred′ (. . ., y)
position.
However, this solution would not be available for three-place non-
causative states. Verbs of envy are possible candidates for this class
(Kailuweit 2012: 118). In line with two-place stative verbs of emotion such
as love or hate, envy takes an emoter argument, the person who experiences
envy, and a target argument, the envied person. In addition, envy possesses
a third argument expressing the object or circumstance the target argu-
ment is envied for.22 For the English verb envy, this third argument could be
realized as a direct core argument (21a) or as an oblique (21b).
(21) a. Many people envied the shareholders their large profits.

b. Max envied Paul for his wealth.
c. Max envied him.
The fact that the third argument, the object of envy, can also be marked
by for (21b) may suggests that it is an adjunct, which gets incorporated into
the core as a direct argument, by analogy with recipient beneficiaries,
such as Max baked a cake for Paul vs. Max baked Paul a cake. Thus, a possible LS for
(21b) could be because.of′ ([have′ (Paul, wealth)], [envy′ (Max, Paul)]). However,
in (21a) and (21b) the target, the envied person, is the undergoer, while the
for-complement becomes the undergoer in the recipient-beneficiary
construction Max baked a cake for Paul ! Max baked Paul a cake.

254 ROLF KAILUWEIT
French envier (‘envy’) shows variable undergoer choice. The target can be
realized either as an undergoer (22a) or as non-macrorole direct core argu-
ment in the dative case (22b). As in English (21c), the object of envy
argument is optional if the target argument is realized as undergoer.
(22) a. Beaucoup d’hommes vous envient pour Madame Linné

Many of men you.2pl.acc envy.3pl for Mrs Linné
‘Many men envy you for Mrs Linné’
(FRANTEXT: HERMARY-VIEILLE, C.)
b. . . .cette belle figure creuse [. . .] il la lui enviait
this pretty face haggard he 3f.sg.acc 3m.sg.dat envied
‘It was this pretty haggard face that he envied him for.’
(FRANTEXT: ETCHERELLI)
The same case variation as in French can be found in German. In add-

ition, the contrast between the two constructions is realized with the help
of the applicative construction with the prefix be-. German neiden (‘envy’)
realizes the target argument as a non-macrorole direct core argument in
the dative case and the object of envy as the undergoer. German beneiden
realizes the target as the undergoer and the object of envy as a prepos-
itional object: jemanden um etwas beneiden (‘to envy somebody for something’).
Hence, the object of envy seems to be the unmarked choice for undergoer
in German.
Admitting that envy is a three-place state would raise the problem of
assigning an LS that provides information about macrorole assignment.
One might consider envy′ (x, y, (z)) as a possible LS for envy. However, the
three argument positions in this LS cannot correspond to the three stative
argument positions in the AUH. As far as the three positions referring to
states are concerned – first argument of pred′ (x, y), second argument of
pred′ (x, y) and argument pred′ (x) – they represent two basically different
types of state. On the one hand, there are two-place states (pred′ (x, y)), for
example verbs of emotion (love′ (x, y)) or possession (have′ (x, y)), with a
less affected second argument (a theme in Gruber’s terms). On the other
hand, there are one-place states (pred′ (x)) with an affected prototypical
undergoer argument, for example resultative verbs of dying (BECOME dead
′ (x)) or destruction (BECOME broken′ (x)). Hence, the structures pred′ (x)
and pred′ (x, y) logically exclude each other and cannot be added to
each other.
Note that verbs of envy are not the only candidates for non-causative
three-place states. A more language-specific case is the French verb en vouloir
à quelqu’un de quelque chose ‘to be angry with someone about something’.23
The argument denoting the person who experiences anger is realized as the
subject, and hence it takes a macrorole. In contrast to verbs of envy, en
vouloir is M-intransitive. Therefore, the two other arguments, the person the
anger is directed at, and the circumstance that the anger is related to, are

realized as a direct core argument in the dative case and as an oblique object
respectively. In comparison with verbs of envy, en vouloir raises an additional
problem: the nature of the macrorole. RRG describes non-episodic verbs of
anger as instances of verbs of internal experience with the LS feel′ (x, y). The
x-argument corresponds to the experiencer, and the y-argument is a
predicate of sensation having one internal argument that is not accessible
for macrorole assignment. Therefore, the predicates of internal sensation
are by default M-intransitive. As intransitive states, they select an undergoer.
The undergoer macrorole is assigned to the x-argument, the experiencer.
(23) Episodic verbs of emotion (Van Valin and LaPolla 1997: 156, 402)
Pat is angry at Kelly.
[feel′ (Pat ¼ experiencer ! 1. arg. of pred′ (x, y) ! Undergoer ! PSA,
[angry.at′ (Kelly)] ¼ sensation ! 2. arg. of. pred′ (x, y))]
A possible LS for three-place en vouloir would be feel′ (x, angry.at′(y).about′

(z)) with two internal arguments of the sensation predicate. However, the
feel′ (experiencer, sensation) analysis is neither convincing for the
French three-place en vouloir nor for its two-place German or Spanish coun-
terparts zürnen and rabiar ‘to be angry at’. While German zürnen realizes the
argument denoting the person the anger is directed at as a dative, Spanish
rabiar expresses this with a preposition (contra, lit. ‘against’). Independently
of these syntactic differences, this argument is obligatory for French en
vouloir, German zürnen and Spanish rabiar. To consider an obligatory argu-
ment inaccessible for macrorole assignment seems to be problematic. In
addition, if the argument denoting the person the anger is directed at were
embedded into the sensation predicate, ‘being angry at James’ and
‘being angry at Paul’ would be two different sensations. Kailuweit (2005b,
2013) proposes that French en vouloir, German zürnen and Spanish rabiar are
atypical activities rather than states. They denote an emotional behaviour
motivated by anger and directed at a person held responsible for the
annoying situation.24 Therefore, the macrorole is an actor and not an under-
goer. An LS that parallels the one proposed for verbs of saying (Staudinger
et al. 2008) would be able to deal with macrorole assignment and linking.
4.5 Alternative Approaches to Macroroles
The fact that languages such as German or Russian possess direct core
arguments in the dative case, which behave differently from other non-
macrorole arguments in syntax, led to intense discussion of the necessity
for a third macrorole. The German passive construction with bekommen
(‘get’) seems to be a candidate for illustrating the theoretical contribution
that a third macrorole could make.
Diedrichsen (2008) discusses the following data.

256 ROLF KAILUWEIT
(24) LS: [do′ (Eltern, Ø)] CAUSE [BECOME have′ (ich, Computer)]
a. Active voice
Meine Eltern haben mir diesen Computer geschenkt
my parents have.3pl 1sg.dat dem.acc computer given
‘My parents gave me this computer.’
b. Passive voice
Dieser Computer ist mir von meinen Eltern geschenkt worden
dem.nom computer be.3sg 1sg.dat by my parents given become
‘This computer was given to me by my parents.’
c. Bekommen-passive voice
Ich habe diesen Computer von meinen Eltern geschenkt bekommen
I.nom have.1sg dem.acc computer by my parents given received
‘I got/received this computer, given by my parents’, ‘I was given this computer
by my parents.’
(25) a. Er bekam von der Polizei den

3m.sg.nom get/receive.3sg.pst by def.f.sg.dat police def.m.sg.acc
Führerschein entzogen.
driving licence.sg revoke.pst
‘He got/received the driving licence revoked by the police.’
b. *Er bekam den Führerschein. Er wurde von der Polizei entzogen.
‘He received the driving licence. It was revoked by the police.’
In addition to the passive construction where the privileged syntactic argu-

ment (PSA) is the argument marked by accusative in the active voice (24b),
German allows the dative-coded recipient argument of many ditransitive
verbs to become a PSA in the passive (24c). PSA assignment to the accusative
argument of the active construction requires the werden-passive construc-
tion, while PSA assignment to the dative argument of the active construc-
tion is realized by the bekommen-passive construction.
Diedrichsen (2008) analyses the bekommen-passive construction in (24c) as a
nuclear juncture, as Bentley (2006: 39) does for the corresponding Italian
construction with avere ‘have’. However, German bekommen is further gram-
maticalized as shown in (25), where it loses its physical reception meaning.
Hence, according to Diedrichsen (2008), the PSA-argument in (25) cannot be
considered the actor of a complex predicate bekommen (‘receive’) þ entziehen
(‘revoke’).
Since PSA choice is restricted to macrorole arguments, Diedrichsen
(2008) considers Lehmann et al.’s (2004) proposal of introducing a third
macrorole, called ‘indirectus’. Like the proto-recipient introduced by
Primus (1999) into the proto-role-approach,25 the indirectus controls and
affects the undergoer, but is controlled and affected by the actor. Lehmann
et al. (2004: 19) define the indirectus as a generalization about the thematic
relations (microroles) recipient, beneficiary, experiencer, emittent
(the participant of a transfer event that gives something away) and sym-
patheticus (the secondary affected participant with a three-place
predicate).

However, the notion of indirectus is restricted to the third argument of

three-place predicates. Hence, from a syntax-to-semantics viewpoint, gener-
alizations concerning the semantics of dative arguments cannot be cap-
tured.26 A third macrorole would be a qualitatively different concept in
comparison with actor and undergoer (Van Valin 2004). While the two
established macroroles are universally valid for the description of morpho-
syntactic phenomena, the third macrorole is not. As Lehmann et al. (2004)
themselves recognize, some languages do not permit three core arguments.
From a semantics-to-syntax viewpoint, there is no consistent treatment of
the arguments that are candidates for the indirectus. German, Russian and
Dyirbal assign dative to the recipient, beneficiary, experiencer, emit-
tent, or sympatheticus of three-place predicates, while English or
Jakaltek marks this argument with an adposition (Van Valin 2004: 81). In
English, the assumption of a third macrorole would single out the to-PP as a
special type of prepositional object. This is not justified on morphosyntactic
grounds.
As far as the German bekommen-passive is concerned, Diedrichsen (2008)
rejects Lehmann et al.’s approach, putting forward two alternative solu-
tions. First, she considers the bekommen-passive as a PSA-selection mechan-
ism whose function is to select a highly topical dative undergoer as PSA. This
would imply accepting datives as undergoers, which is not foreseen in
standard RRG. Alternatively, an approach more in line with standard RRG
would account for the bekommen construction as a lexical passive in that
undergoer status is assigned to a dative recipient that is the non-macrorole
argument of the ditransitive active.
From the viewpoint of typology, Bickel (2011) and Bickel et al. (2014) refer
explicitly to the concept of macroroles to develop a descriptive toolkit for
the different alignment types of ditransitive constructions. Bickel et al.
distinguish between different sets of macroroles (or proto-roles) for one-
place, two-place and three-place predicates. Building on a tradition in typ-
ology which started with Comrie (1978) and Dixon (1979), they distinguish
between S as the argument of an intransitive verb, A as the most actor-like
argument in a transitive verb, and O or P (in Bickel et al. 2014)27 as the not
most actor-like argument in a transitive verb. In addition, they label the two
non-actor-like arguments of ditransitives as G (for goal) and X or T (for
theme). The generalized semantic roles G28 and X/T stand for the most
goal-like and for the other (not most goal-like, not most actor-like) argument
respectively. While the only semantic motivation for S is the fact that the
predicate takes only one argument, S does not serve to describe split
intransitivity. A and O/P seem to correspond to actor and undergoer in
RRG. However, the generalized semantic roles are assigned to any two-place
predicate, even to predicates lacking a referential second direct core argu-
ment. In addition, neither G nor X, the two less active arguments of three-
place predicates, have to be direct core arguments. In typology, the function
of the generalized semantic roles is to serve as a tertium comparationis for

258 ROLF KAILUWEIT
Figure 4.2 Four types of languages, accusative and ergative alignment (Haspelmath 2008)
different forms of alignment. For example, marking S and A in the same way
characterizes accusative alignment. While grouping O/P and X/T together
corresponds to indirective alignment, secundative alignment is character-
ized by marking O/P and G/R identically (Bickel 2011).
Haspelmath (2008) pursues a very similar approach, which he includes
directly into the RRG framework. He proposes four macroroles: A ¼ actor,
U ¼ undergoer of monotransitive sentences, R ¼ macro-recipient, corres-
ponding to Primus’s (1999) proto-recipient and the aforementioned indir-
ectus of Lehmann et al. (2004), and finally T ¼ (macro-)theme (see
Figure 4.2). These macroroles are defined in purely semantic terms. They
account for the parallels between monotransitive and ditransitive
alignments in a straightforward way without needing to refer to marked
or unmarked undergoer choice, as in standard RRG, or to parameterize
undergoer selection for the indirective–secundative contrast, as in Guer-
rero and Van Valin (2004).
Although U, T, R and A seem to correspond to the four rightmost pos-
itions of the AUH, it is unclear how their semantic values could be deter-
mined in a precise way. Haspelmath (2011) himself criticizes the approach
of Bickel et al. (2014), but his critique refers indirectly to his own RRG
approach, too (Haspelmath, p.c., 30 January 2013). His point is that Bickel’s
generalized semantic roles are universal semantic categories aimed at the
description of individual languages. However, being not only syntactically
undetermined but also semantically highly heterogeneous, they do not
serve to draw typological generalizations. Indeed, it remains unclear in
both approaches what agent-like, goal-like or theme-like mean beyond
purely intuitive notions.29

Table 4.3 Features and values for activity clusters (from Kailuweit 2013)
þc c c þm m m r r þr
4 2 0 2 1 0 0 2 4
In line with Comrie’s approach, Haspelmath (2011) points out that S, A, P/O/
U, T/X, and R/G should only be comparative concepts referring to the syntactic
relations associated in a given language with verbs selecting a prototypical
agent and patient (A:P), verbs of transfer (A:R:T), and one-place verbs of
uncontrolled change of state such as verbs of dying (S). This means that not
every two-place predicate assigns an active and a passive generalized semantic
role. Only predicates that code the opposition of the arguments, like proto-
typical verbs with an agent and a patient, do: for example verbs like English
see, like, own or enter follow the pattern of break or kill, but appeal to and belong
to do not. This is in line with standard RRG. However, Haspelmath’s (2011)
comparative concepts do not apply to all predicates in every language and do
not formulate hypotheses about universal semantic structure. Hence, they
differ considerably from RRG macroroles.
Kailuweit (2013, 2018) maintains the two macroroles of standard RRG,
devising, however, a feature-based Activity Hierarchy, which departs from
the RRG AUH. Following Rozwadowska (1988), Kailuweit (2013) takes three
features into account: causative and/or control [c], mental (sentient) [m],
and resultative (change of state) [r]. In line with Reinhart (2002), these
features can assume one of the three values þ, or . In addition, the
features are weighted (see Table 4.3). The feature [c] is a strong actor
feature, [m] is a weak actor feature, and [r] is a strong undergoer feature.
The presence of a strong feature, indicated with [þ], will duplicate the
value of the presence of a weak feature. If an argument is underspecified
for one feature or has an intermediate degree of that feature [], the value
will be half of the [þ] value.
The combination [þcþmr] represents the prototypical actor with the
value 4þ2þ0 ¼ 6, while the prototypical undergoer corresponds to the
combination [cmþr] 0þ04 ¼ 4. Nine intermediate combinations are
mathematically possible. Hence, the system distinguishes eleven possible
degrees of activity. The attribution of the values follows from a detailed
semantic analysis of events and participants, evaluating aspectual proper-
ties as well as causativity, sentience, responsibility and control.
In the following, sketch-like descriptions will be given to account for the
non-causative three-place predicates talk and envy as well as for some two-
place verbs of emotion. Some of the examples discussed above are repeated
for convenience.
(26) Bob talked to Pat about Sandy
(27) a. Many people envied the shareholders their large profits

b. Max envied Paul for his rewarding task

260 ROLF KAILUWEIT
(28) a. Bill’s owning a gun frightened Martha

b. The mother feared for her children
c. The mother feared her children
As far as talk is concerned, the speaker controls their activity to a

certain degree, but not entirely (mimics and gesture, unforeseen effects
of the speech, slips of the tongue, etc.), hence, the value for [c] is . The
addressee does not control the event but takes part in it as a sentient
being. The topic of conversation neither controls the event nor is it
affected by it. Hence, the speaker is the most active argument assuming
the actor macrorole. The addressee is neither the most active nor the
most passive argument. As with other arguments exhibiting an intermedi-
ate degree of activity, it is realized with the preposition to in English (and
the dative in the case of French parler ‘talk’). The topic of conversation
is the most passive argument of the event, although it assumes rather a
high degree of activity (0 in comparison with 4 of the prototypical
patient). This may explain the realization with a preposition (about/of in
English, sur/de in French).
Verbs of envy denote non-episodic emotions with three participants. In
episodic emotions, prototypically the target, for example the person or
the state of affairs the emotion is directed at, is at the same time the
causer of the emotion. The experiencer is affected, undergoing a change
of emotional state in a given situation. This is the case of frighten in
(28a) selecting a target that functions as an inanimate causer. By
contrast, emoters of non-episodic emotions express a subjective judge-
ment that they are held responsible for, because they control it to a
certain degree [c]. Some verbs of emotion show another argument,
helping to identify the target. In (28b), for example, the children are in
danger, but they are not the danger themselves as in (28c). Hence, the
children represent a point of reference to find out about the danger as
the target of the mother’s fear. With envy, the target and the point of
reference are realized. The envy is directed to a person being the target.
The third argument specifies why this person has become the target of
envy. The attribution of [m] accounts for the fact that the point of
reference is a human being (prototypically, but not necessarily with fear
for) or a property associated with a human being (in the case of envy). For
English envy, the emoter as the most active argument represents the actor,
the target as the most passive argument the undergoer. The point of
reference of a middle degree of activity appears as the non-macrorole
argument.30
Further research will determine whether the Activity Hierarchy can
describe the same number and even more predicate classes in as coherent
a way as the LS-based AUH does.31 Perhaps LSs and activity values can also be
combined in the lexical entries of the predicates to enable a fine-grained
semantic description of the linking-relevant properties.

Table 4.4 Activity Hierarchy of arguments of selected predicates
[þcþmr] 6 AGENT - CAUSER murder (x,. . .)

[þcmr] 4 INANIMATE CAUSER frighten (x,. . .)
[cþmr] 4 SPEAKER / EMOTER talk (x,. . .), envy (x,. . .), fear (x,. . .),
fear for (x,. . .)
[cþmr] 2 ADDRESSEE talk (. . .,y,. . .)
[cmr] 1 POINT OF REFERENCE OF EMOTIONS fear for (. . .,y), envy (. . .,y,. . .)
[cmr] 0 TARGET OF EMOTIONS / TOPIC OF CONVERSATION envy (. . .,z), talk (. . .,z)
[cþmþr] -2 AFFECTED EXPERIENCER frighten (. . .,y)
[cmþr] -4 PATIENT (undergoing a change of state) murder (. . .,y)
4.6 Conclusion
The macroroles of RRG constitute a powerful tool to address the challenges

of argument realization. Specifically, the theory of macrorole assignment
captures the key insights of the approaches called ‘aspectual’, ‘causal’ and
‘neo-constructionist’ by Rappaport Hovav and Levin (2015). In the vast
majority of cases, RRG makes correct predictions about argument linking
by using predicate decompositions and identifying the relevant facets of
the predicate’s meaning. Individual facets may be more effectively cap-
tured by other approaches, which are mainly, or exclusively, ‘feature-
based’, ‘decompositional’, ‘aspectual’, ‘causal’ or ‘neo-constructionist’.
Nevertheless, macrorole assignment in RRG currently seem to be the most
complete approach to capture meaning generalizations relevant to the
syntactic realization of arguments.
References
Abreu Gomes, Cristina. 2003. Dative alternation in Brazilian Portuguese:

Typology and constraints. Language Design 5: 67–78.
Anderson, Stephen. 1971. On the role of deep structure in semantic inter-
pretation. Foundations of Language 6: 197–219.
Bellosta von Colbe, Valeriano. 2004. Das indirekte Objekt als syntaktisches
Argument ohne Makrorolle. In Kailuweit and Hummel (eds.), 183–204.
Bentley, Delia. 2006. Split Intransitivity in Italian. Berlin: De Gruyter.
Bickel, Balthasar. 2011. Grammatical relations typology. In Jae Jung Song
(ed.), The Oxford Handbook of Linguistic Typology, 399–444. Oxford: Oxford
University Press.
Bickel, Balthasar, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-
Makarevich. 2014. Semantic role clustering: An empirical assessment of
semantic role types in non-default case assignment. Studies in Language
38(3): 485–511.
Borer, Hagit. 1994. On the projection of arguments. In Elena Benedicto and
Jeffrey Runner (eds.), Functional Projections, 19–47. Amherst: GSLA.

262 ROLF KAILUWEIT
Borer, Hagit. 2005. Structuring Sense II: The Normal Course of Events. Oxford:
Comrie, Bernard. 1978. Ergativity. In Winfred P. Lehmann (ed.), Syntactic
Typology: Studies in the Phenomenology of Language, 329–394. Austin: Univer-
sity of Texas Press.
Diedrichsen, Elke. 2008. The grammaticalization of the bekommen-passive in
a RRG-perspective. In Kailuweit et al. (eds.), 87–145.
Dixon, Robert M. W. 1979. Ergativity. Language 55: 59–138.
Dowty, David. 1979. Word Meaning and Montague Grammar: The Semantics of Verbs
and Times in Generative Semantics and in Montague’s PTQ. Dordrecht: Reidel.
Dowty, David. 1991. Thematic proto-roles and argument selection. Language
67(3): 547–619.
Dryer, Matthew S. 1986. Primary objects, secondary objects, and antidative.
Language 62: 808–845.
Fillmore, Charles J. 1968. The case for case. In Emmon Bach and Robert
and Winston.
Fillmore, Charles J. 2003. Valency and semantic roles: The concept of deep
structure case. In Vilmos Ágel, Ludwig M. Eichinger, Hans Werner Eroms,
Peter Hellwig, Hans Jürgen Heringer and Henning Lobin (eds.), Dependenz
und Valenz: Ein internationales Handbuch der zeitgenössischen Forschung /
Dependency and Valency: An International Handbook of Contemporary Research
1, 457–475. Berlin: Walter de Gruyter.
François, Jacques. 1997. La place de l’aspect et de la participation dans les
classements conceptuels des prédications verbales. In Jacques François and
Guy Denhière (eds.), Sémantique linguistique et psychologie cognitive. Aspects
théoriques et expérimentaux, 119–156. Grenoble: Presse Universitaire.
Grimshaw, Jane. 1990. Argument Structure. Cambridge, MA: MIT Press.
Gruber, Jeffrey S. 1965 [1976]. Studies in Lexical Relations. PhD dissertation,
MIT. Published as: Lexical Structures in Syntax and Semantics. Amsterdam:
North-Holland.
Guerrero, Lilián and Robert D. Van Valin, Jr. 2004. Yaqui and the analysis of
primary object Languages. International Journal of American Linguistics 70(3):
290–319.
Haspelmath, Martin. 2008. Ditransitive constructions: Towards a new Role
and Reference Grammar account? In Van Valin (ed.), 75–100.
Haspelmath, Martin. 2011. On S, A, P, T, and R as comparative concepts for
alignment typology. Linguistic Typology 15: 535–689.
Hockett, Charles. 1958. A Course in Modern Linguistics. New York: Macmillan.
Ibáñez Cerda, Sergio. 2008. ‘Saying’ verbs in Spanish. Deepening the lexical
semantics description. In Van Valin (ed.), 3–21.
Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cam-
bridge, MA: MIT Press.

Jackendoff, Ray. 2007. Language, Consciousness, Culture: Essays on Mental Struc-

ture (Jean Nicod Lectures). Cambridge, MA: MIT Press.
Kailuweit, Rolf. 2004. Protorollen und Makrorollen. In Kailuweit and Hum-
mel (eds.), 84–104.
Kailuweit, Rolf. 2005a. Lokativalternanz bei transitiven Verben – Englisch,
Französisch, Spanisch und Deutsch im Vergleich. In Barbara Wotjak
and Christian Schmitt (eds.), Beiträge zum romanisch-deutschen und inner-
romanischen Sprachvergleich, Vol. 2, 183–196. Bonn: Romanistischer
Verlag.
Kailuweit, Rolf. 2005b. Linking: Syntax und Semantik französischer und italie-
nischer Gefühlsverben. Tübingen: Niemeyer (Linguistische Arbeiten).
Kailuweit, Rolf. 2008. A RRG description of locative alternation verbs in
English, French, German and Italian. In Kailuweit, Wiemer, Staudinger
and Matasović (eds.), 328–355.
Kailuweit, Rolf. 2012. Macropapeles: entre semántica y sintaxis. In Carlos
González Vergara, Lilián Guerrero and Ricardo Mairal (eds.), El funcionalismo
en la teoría lingüística: La Gramática del Papel y la Referencia, 103–123. Tres
Cantos: Ediciones Akal.
Kailuweit, Rolf. 2013. Radical Role and Reference Grammar (RRRG):
A sketch for remodelling the syntax–semantics interface. In Brian
Nolan and Elke Diedrichsen (eds.), Linking Constructions into Functional
Linguistics: The Role of Constructions in Grammar, 103–41. Amsterdam: John
Benjamins.
Kailuweit, Rolf. 2018. Activity Hierarchy and argument realization in (R)RRG.
In Rolf Kailuweit, Lisann Künkel and Eva Staudinger (eds.), Applying and
Expanding Role and Reference Grammar, 189–213. Freiburg: Albert-Ludwigs-
Universität, Universitätsbibliothek.
Kailuweit, Rolf and Martin Hummel (eds.). 2004. Semantische Rollen. Tübingen:
Narr.
Kailuweit, Rolf, Björn Wiemer, Eva Staudinger and Ranko Matasović (eds.).
2008. New Applications of Role and Reference Grammar. Newcastle upon Tyne:
Cambridge Scholars Publishing.
Kibrik, Alexander E. 1985. Toward a typology of ergativity. In Johanna
Nichols and Anthony Woodbury (eds.), Grammar Inside and Outside the
Clause, 268–323. Cambridge: Cambridge University Press.
Kibrik, Alexander E. 1997. Beyond subjects and objects: Towards a compre-
hensive relational typology. Linguistic Typology 1: 279–346.
Langacker, R. W. 1990. Concept, Image, and Symbol. Berlin: Mouton de Gruyter.
Langacker, R. W. 1991. Foundations of Cognitive Grammar. 2: Descriptive Applica-
tion. Stanford, CA: Stanford University Press.
Lehmann, Christian, Yong-Min Shin and Elisabeth Verhoeven. 2004. Direkte
und indirekte Partizipation. Zur Typologie der sprachlichen Repräsentation konzep-
tueller Relationen. Munich: Lincom Studies in Language Typology.
Levin, Beth and Malka Rappaport Hovav. 1995. Unaccusativity at the Syntax–
Lexical Semantics Interface. Cambridge, MA: MIT Press.

264 ROLF KAILUWEIT
Nissenbaum, Helen Fay. 1985. Emotion and Focus. Stanford, CA: CSLI
Publications.
Pinker, Steven. 1989. Learnability and Cognition: The acquisition of Argument
Structure. Cambridge, MA: MIT Press.
Primus, Beatrice. 1999. Case and Thematic Roles: Ergative, Accusative, Active.
Tübingen: Niemeyer.
Rappaport Hovav, Malka and Beth Levin. 1988. What to do with theta roles.
In Wilkins (ed.), 7–36.
Rappaport Hovav, Malka and Beth Levin. 2015. The syntax–semantics inter-
face. In Shalom Lappin and Chris Fox (eds.), The Handbook of Contemporary
Semantic Theory (2nd ed.), 593–624. Oxford: Wiley-Blackwell.
Rauh, Gisa. 1988. Tiefenkasus, thematische Relationen und Thetarollen. Tübingen:
Narr.
Reinhart, Tanya. 2002. The Theta system: An overview. Theoretical Linguistics
28(3): 229–290.
Rozwadowska, Bozena. 1988. Thematic restrictions on derived nominals. In
Wilkins (ed.), 147–165.
Ruwet, Nicolas. 1972. Théorie syntaxique et syntaxe du français. Paris: Seuil.
Staudinger, Eva, Matthias Hartung and Rolf Kailuweit. 2008. Linking syntax
to semantics: template selection and PP-Attachment ambiguities. In Kai-
luweit, Wiemer, Staudinger and Matasović (eds.), 389–413.
Tenny, Carol. 1994. Aspectual Roles and the Syntax–Semantics Interface. Dor-
drecht: Kluwer.
Tesnière, Lucien. [1959] 1965. Eléments de syntaxe structurale. Paris:
Klincksiek.
Robert D. Van Valin Jr. (ed.), Advances in Role and Reference Grammar, 1–164.
Van Valin, Robert D., Jr. 1999. Generalized semantic roles and the syntax–
semantics interface. In F. Corblin, C. Dobrovie-Sorin and J.-M. Marandin
(eds.), Empirical Issues in Formal Syntax and Semantics, Vol. 2, 373–389. The
Hague: Thesus.
Van Valin, Robert D., Jr. 2004. Semantic macroroles in Role and Reference
Grammar. In Rolf Kailuweit and Martin Hummel (eds.), Semantische Rollen,
62–82. Tübingen: Narr.
Van Valin, Robert D, Jr. 2006. Semantic macroroles and language processing.
In I. Bornkessel et al. (eds.), Semantic Role Universals and Argument Linking:
Theoretical, Typological and Psycholinguistic Perspectives, 263–302. Berlin: Mou-
ton de Gruyter.
Van Valin, Robert D., Jr. 2007. The Role and Reference Grammar analysis of
three-place predicates. Suvremena Lingvistika 63(1): 31–64.

Van Valin, Robert D., Jr. 2010. Role and Reference Grammar as a framework
for linguistic analysis. In Bernd Heine and Heiko Narrog (eds.), The Oxford
Handbook of Linguistic Analysis, 703–738. Oxford: Oxford University Press.
Van Valin, Robert D., Jr. 2013. Lexical representation, co-composition, and
Van Valin, Robert D., Jr. 2018. Some issues regarding (active) accomplish-
and Expanding Role and Reference Grammar, 5–26. Freiburg: Freiburg Univer-
sity Library.
Van Valin, Robert D., Jr. and Randy R. LaPolla. 1997. Syntax: Structure, Meaning
Van Valin, Robert. D., Jr. and David Wilkins 1996. The case for ‘effector’: Case
roles, agents, and agency revisited. In Masayoshi Shibatani and Sarah. A.
Thompson (eds.), Grammatical Constructions: Their Form and Meaning,
289–321. Oxford: Oxford University Press.
Wilkins, Wendy (ed.). 1988. Thematic Relations, Vol. 21: Syntax and Semantics.
New York: Academic Press.
Notes
1 The expression ‘generalized semantic roles’ has been established as a

neutral term to refer to the RRG macroroles, proto-roles (Dowty 1991),
Kibrik’s (1985, 1997) hyperroles, and Langacker’s (1990, 1991) role arche-
types, etc. (see Van Valin 1999; Kailuweit 2004; Rappaport Hovav and
Levin 2015).
2 Tesnière ([1959] 1965) calls the prepositional phrase of the passive con-
struction a ‘counter-agent’, but he fails to capture the semantics of the
subject of the passive construction.
3 According to Levin and Rappaport Hovav (1995), in the case of intransitive
verbs, telicity and directed change are criteria for unaccusativity.
4 ‘The telicity of the event depends on whether the patient of the event
shows a bounded or unbounded change in the gradable property associ-
ated with the verb’ (Rappaport Hovav and Levin 2015: 607).
5 Following the prevailing view at that time that all accomplishments are
causative and that all causatives are accomplishments (Dowty 1979: 186),
Grimshaw (1990) considered causativity an aspectual property. However,
later work in RRG (Van Valin and LaPolla 1997: 97) has shown that
causativity is independent of Aktionsart (see Section 4.4.1).
6 Logical structures in RRG are based on Dowty’s (1979) formalism. As will
become clear when causativity is discussed, the details of Aktionsart-
driven verb classification were later developed in RRG in ways which
diverge from Dowty (1979).

266 ROLF KAILUWEIT
7 However, even in later versions of RRG the concept of macrorole is

justified by the assumption that each macrorole ‘subsumes a number
of specific argument types (thematic relations). The generalized agent-
type role will be termed actor and the generalized patient-type role will
be called undergoer’ (Van Valin and LaPolla 1997: 141; see also Van Valin
2006: 60).
8 ‘Actor and undergoer are generalizations across classes of specific
argument positions in logical structure’ (Van Valin and LaPolla
1997: 142).
9 For mnemotechnical purposes, the first argument of kill is not named
agent but effector. The latter is a generalized label for the (do′ (x,. . .))
position (Van Valin and Wilkins 1996).
10 There are isolated cases of non-causative achievements and accomplish-
ments with a do′-component in their LS. INGR or BECOME plus activity
predicate describe an onset of an action, e.g. Russian zaplakat’ ‘burst out
crying’ (INGR do′ (x, [cry′ (x)])) vs. zagovorit’ ‘start talking’ (BECOME do′ (x,
[talk′(x)])) (Van Valin and LaPolla 1997: 104).
11 Van Valin and LaPolla (1997: 143) expresses this idea referring to the-
matic relations and not to the positions on the hierarchy they stand for:
‘Actor and undergoer are generalized semantic roles whose prototypes
are the thematic relations AGENT and PATIENT, respectively’ (see also
Van Valin 2004: 63; Van Valin 2005: 60).
12 This is not only the case with three-place constructions such as to give
somebody something, but also with two-place activities such as to eat fish.
See Van Valin and LaPolla for one such case, Italian mangiare (‘eat’) (1997:
148–150).
13 In this example, the symbol Ø refers to a unspecific action that can
cause death.
14 The so-called dative alternation does not appear in German or in the
standard varieties of Romance languages (but see Abreu Gomes (2003) for
colloquial Brazilian Portuguese). Kailuweit (2005a) showed that Pinker’s
(1989) semantic subclass approach to locative alternation does not hold
for German or Romance languages. Some of the English subclasses that,
according to Pinker, allow for locative alternation do not have any
instances in French, Spanish or German, while other classes that do
not permit alternation in English do so in German, French or Spanish.
However, in every alternating class in these languages, we also find verbs
that do not alternate. Hence, there is no uniform relation between
Pinker’s semantic classes of locative verbs and locative alternation (see
Kailuweit 2005a, 2008).
15 In a primary object language the recipient argument of a ditransitive
verb – the y-argument of BECOME be.at’(y,z) in the LS of these verbs – is
treated like the patient of a monotransitive verb. The theme argument –
the z-argument of BECOME be.at’(y,z) – is marked in a distinct way (Dryer
1986).

16 Van Valin (2004: 74–78) pointed out that macroroles are basically seman-
tic categories that play a role in syntax. Syntax is considered non-
autonomous in RRG. Therefore, syntactic (in)transitivity is at least partly
semantically motivated.
17 Primus (1999: 36) substitutes Dowty’s (1991) proto-agent property ‘vol-
ition’ by ‘control’ to describe cases that exclude ‘volition’ in the strict
sense, for example responsibility or the mere capacity to start or stop an
event. In RRG, too, responsibility is referred to, at least informally, to
define the properties of the actor: ‘the actor is the entity to which
responsibility for the action or event is attributed’ (Van Valin and LaPolla
1997: 143).
18 ‘Seeing, thinking, liking, believing, etc., involve some kind of internal
activity (mental, emotional or perceptual) on the part of the participant,
whereas being seen, being thought about, being liked or being believed
does not require any action or effort of any kind on the part of the
participant. Hence the participant denoted by the first argument is more
active and hence more agent-like than the participant referred to by the
second argument, and, accordingly, the first argument is closer to the
agent end of the hierarchy than the second argument’ (Van Valin
2005: 58).
19 The existence of non-causative three-place predicates is widely ignored in
the literature. Haspelmath (2008: 14) refers at least incidentally to talk as
a verb ‘lacking an undergoer’, adding that ‘intuitively it is strange to
claim that ditransitive constructions are intransitive’.
20 Ibáñez Cerda (2008: 6) criticizes this representation, highlighting the fact
that the α variable could be realized as a direct object with verbs of saying.
21 Two other classes of two-place activities are verbs of consumption and
creation with an unspecified object: eat pizza, write letters. Note that the
objects of these predicates are not affected undergoers.
22 ‘Envy defines three central roles, that of the envier, the envied, and a
feature or possession of the one envied, over which he is envied’ (Nissen-
baum 1985: 108).
23 The clitic en is a desemantized part of the lexical realization of the
predicate that could be interpreted as a generic placeholder for the state
of affairs (the correlate) the anger is about. Curiously, the de-comple-
ment that can normally be substituted by the clitic en appears alongside
the clitic in the sentence. It is obviously not a fourth argument, but a
coreferential concretization of the correlate.
24 Spanish rabiar is somewhat different. It does not only accept a person as
the second argument but also an object. Hence, it corresponds to English
rage against.
25 For a discussion of Primus’s approach, and its compatibility with the
RRG framework, see Bellosta von Colbe (2004).
26 Bellosta von Colbe (2004) uses the same argument to criticize Primus’s
proto-recipient approach.

268 ROLF KAILUWEIT
27 Bickel et al. (2014) use P (for patient) instead of O and T (for theme)
instead of X. See Haspelmath (2011) for an overview on the history of
denominations for generalized roles in typology.
28 Haspelmath (2011) and other authors refer to G as R (for recipient).
29 Bickel et al. (2014) refer to entailments in line with Dowty (1991), but
Dowty’s approach allowing 1,024 theoretically possible combinations
of proto-role features turns out to be a rather inaccurate tool to deter-
mine the activity degree of an argument in comparison with the AUH.
30 Note that French envier and German neiden/beneiden allow for flexible
undergoer selection, i.e. a construction selecting the second passive
argument as undergoer (secondary alignment in a language of indirec-
tive alignment). Kailuweit’s (2013) approach does not account for
this alternation.
31 Kailuweit (2018) discusses further classes of predicates, especially verbs
of directed motion such as climb, enter or reach.

5
Grammatical Relations
Randy J. LaPolla
5.1 Introduction
This chapter discusses one of the major advances of Role and Reference
Grammar (RRG) relative to other theories of grammar: the view of grammat-
ical relations as construction-based, and so not only language-specific but
construction-specific, rather than being global categories of the whole lan-
guage and found in every language. We also discuss the RRG conception of
the function of grammatical relations in referent tracking, which was one of
the major insights that led to the development (and naming1) of RRG. These
two insights have influenced the development of ideas outside RRG.
5.2 Background
Before the mid 1970s, a common assumption among linguists was that there
is a global category in all languages called ‘subject’ as well as other gram-
matical relations that we can talk about, and most theories assumed some
conception of grammatical relations, though there was much disagreement
about and no universal notion of ‘subject’, the grammatical relation dis-
cussed the most (Platt 1971; Van Valin 1977, 1981; Foley and Van Valin 1977,
1984; Keenan 1976; Gary and Keenan 1977; Comrie 1981). The word ‘subject’
derives from a Latin translation (subiectum) of Greek hypokeímenon ‘the under-
lying thing’, a concept that began with Aristotle’s theory of truth, where
Aristotle defined ‘subject’ as the entity that can have a predication about it,
that is, what the proposition is about, the topic about which a predication is
made. Aristotle did not have a separate term for grammatical subject. This
led to centuries of debate about the nature of subject (see Seuren 1998,
pp. 120–133, for an overview). Attempts were made to distinguish grammat-
ical subject from psychological subject (e.g. von der Gabelentz 1869: 378),
the latter essentially topic, and what was called ‘theme’ in the Prague School

270 RANDY J. LAPOLLA
terminology. A third term, ‘logical subject’ (often now seen as agent), was
sometimes used, but different scholars associated it either with grammat-
ical subject or with psychological subject (particularly in logic). Bloomfield
(1914: 61) used the term ‘subject’ to refer to topics and also to heads
of phrases.
Starting with Van Valin (1977, 1981) and Foley and Van Valin (1977,
1984), there were challenges to the notion of ‘subject’ and other gram-
matical relations (‘direct object’, ‘indirect object’) as global categories
within a single language, and as valid categories cross-linguistically (see
also Dryer 1997). Currently there are three major positions on this ques-
tion: (1) grammatical relations are global within a language and universal
cross-linguistically, and just need to be identified in different languages
(the rationalist/generativist/Chomskyan tradition); (2) grammatical rela-
tions exist, but are not necessarily global and not universal, and so
need to be defined in each language in terms of the constructions that
manifest them, if there are any (most empiricist/typological/explanatory
approaches); and (3) there are no grammatical relations, only part–whole
relations within constructions (Radical Construction Grammar; Croft
2001, 2013). Marantz (1982, 1984) has argued that grammatical relations
should not be seen as primitives or tied to semantic roles. For example,
‘subject’, as a grammatical category, is not simply a particular semantic
role, such as agent (see also Jespersen 1909–1949, vol. III, 11.1). ‘Subject’ is
also not simply topic; it must have grammatical properties beyond just
being what the clause is about. Empiricist linguists would generally agree
with this position.
5.3 The RRG View of Grammatical Relations
The RRG view of grammatical relations is of the second type mentioned

above, though it does not accept the traditional typology of grammatical
relations as ‘subject’, ‘direct object’ and ‘indirect object’. Grammatical rela-
tions are seen as construction-specific conventionalized patterns where the
construction limits the possible interpretations of the role of a particular
participant in the action described in the clause (see Van Valin and LaPolla
1997: 242–316; Van Valin 2005; LaPolla 2006). It is the identification of the
semantic and/or pragmatic role of the referent in an event or state of affairs
that is relevant to the concept of grammatical relations. There are other
types of conventionalized constraints on referent identification in some
languages, such as the gender or noun class markers in many languages,
and the sortal classifiers of Thai and Chinese, but while they do participate
in referent tracking (see Van Valin 1987 for a typology of referent tracking
systems), as these are not relational and do not constrain the interpretation
of the role of the referent in the event, they are not considered relevant to
grammatical relations.

Grammatical Relations 271
Grammatical relations may be conventionalized associations of the pos-

ition of a referring expression in the clause with some semantic role or
macrorole, such as in English, where an immediate preverbal reference to
some referent in a clause with an active transitive verb will constrain the
interpretation to one in which that referent is seen as the Actor2 of the
action denoted by the verb, and an immediate postverbal reference to
some referent in the same clause will constrain the interpretation to one
in which that referent is seen as the Undergoer of the action (e.g. given the
expression Bob hit Bill, the conventions of English usage constrain the
interpretation to one in which Bob must be understood as the one doing
the hitting and Bill must be understood as the one being hit). This is
construction-specific and language-specific, that is, not all constructions
in English work that way and not all languages have the same constraints
on interpretation.
Grammatical relations may be conventionalized associations of marking
on nouns or pronouns with particular semantic roles, such as in English,
where Nominative case marking of a pronoun in an active transitive clause
constrains the interpretation to one in which the referent of the pronoun
is seen as the Actor of the action denoted by the verb, and Accusative case
marking of a pronoun constrains the interpretation to one in which the
referent of the pronoun is seen as the patient or recipient of the action,
such as in HeNominative took meAccusative to the station. In the case of Modern
English, the case marking is largely redundant, in that grammatical rela-
tions are also marked by word order, and in fact case marking has largely
come to be determined by word order, but this was not the case in older
varieties of English, which did not use word order consistently to mark
grammatical relations, and in many other languages, such as Dyirbal (see
below). In these languages the case marking is very important not only for
tracking the roles of referents in discourse, but also for tracking the
different constituent parts of phrases, as they do not necessarily appear
together in the clause.
Grammatical relations may also be conventionalized assumptions that
referring expressions in two clauses both refer to the same referent, such
as in English, where there is a conventionalized assumption of coreference
in conjoined clauses such that a referring expression representing a particu-
lar role in one of the clauses and a particular role implied in the other clause
must be understood as coreferential (e.g. in Jim picked up the newspaper and
threw it, there is a forced assumption that the referent of Jim is the same
referent as the implied Actor of the second clause, the one that threw the
newspaper).3 Many other possible ways of constraining this particular func-
tional domain exist as well. See Section 5.3.3 for more discussion.
Each of these conventionalized forms or constructions has the function of
limiting the possible interpretation(s) of the role of a referent referred to or
implied in an utterance, to aid in the interpretation of the identity and role
of the referent. Although traditionally these different constructions have

been seen as part of one grammatical category (e.g. ‘subject’), they are not
one category, but instead are individual ways of constraining the interpret-
ation of who is doing what to whom, and languages differ in terms of
whether or not they constrain this functional domain at all, and if they do
constrain it, they differ in terms of which particular roles are identified, and
the particular mechanisms used to constrain the interpretation.
5.3.1 Privileged Syntactic Argument: Controller and Pivot

As the phenomenon we are talking about is construction-specific, and there
are many different types of restricted neutralizations, the term ‘subject’ is
not appropriate, and instead we use the term ‘privileged syntactic
argument’ (PSA) for an argument that is the controller or pivot of a
restricted neutralization of semantic roles for grammatical purposes, gener-
ally referent tracking (i.e. keeping track of who is doing what in discourse).
No other grammatical relations are recognized in this approach; the char-
acteristics that have traditionally been associated with ‘direct object’, such
as taking accusative case, being able to appear as the privileged syntactic
argument in a passive clause or being the target of applicative construc-
tions, are seen as properties of the macrorole Undergoer, while those associ-
ated with ‘indirect object’ are seen as properties of the non-macrorole direct
core argument (see the discussion of (5) below).
The terms ‘controller’ and ‘pivot’ (first used in Heath 1975) refer to differ-
ent types of PSA, as in the examples in (1), the English conjoined clause
coreference construction, where the immediately preverbal (core-initial)
referring expression in the first clause is the controller, while the implied
argument of the second clause is the pivot of the construction (marked by
‘(pivot)’, where the implied argument would be if it appeared in an inde-
pendent clause), regardless of whether either clause is an active or passive
construction.
(1) a. Bobcontroller handed Jim the money and (pivot) left. [A,S]
b. Jimcontroller was handed the money by Bob and (pivot) left. [S,S]
c. The moneycontroller was handed to Jim but (pivot) not seen again
after that. [S,S]
d. Jimcontroller took the money and (pivot) was seen later buying a
new car. [A,S]
e. Jimcontroller took the money and (pivot) thanked Bob for it. [A,A]
f. Jimcontroller smiled and (pivot) took the money. [S,A]
In this construction, there is obligatory coreference between the controller,

whether it is the Actor (A) of an active transitive clause or single direct
argument of an intransitive clause (S)4 or the Undergoer of a passive clause
(S), and the pivot, again whether it is A or S. The construction then aids in
the inference of who is doing what. The restricted neutralization we find in
this particular construction is [A,S], but as shown in the examples, the
correspondence can be [A,A] or [S,S] as well, as what is important is that

the arguments participating in the coreference are grammatically either

A or S; the Undergoer of an active transitive clause does not participate in
this coreference. The Undergoer argument cannot simply be left to implica-
tion when using this structure, and so a different construction must be used
to allow the Undergoer to participate in the coreference of the conjoined
clause coreference construction. The construction used is the English pas-
sive construction, as in (1b, c ,d), as it is an intransitive clause in which the
Undergoer is the PSA. Using the passive construction in this conjoined
clause coreference construction allows the Undergoer argument (regardless
of what specific semantic role it has) to participate in the coreference
construction by casting it as an S.
The English passive construction allows limited variable access to the
syntactic controller and syntactic pivot positions. Without that construc-
tion the Undergoer would not be able to appear as the PSA. Some languages
manifest constructions with a particular restricted neutralization but do
not have constructions that allow variable access to the PSA (i.e. they do
not have alternative voice constructions such as passive; e.g. Enga (Papua
New Guinea), Warlpiri (Australia), Lakhota (North America); see Van Valin
and LaPolla 1997, pp. 274–285 for details). These constructions are said to
have an invariable syntactic controller and invariable syntactic pivot. Some
languages have one or a few constructions (e.g. the English passive con-
struction or the Dyirbal antipassive construction – see the next section)
that allow limited variable access to the PSA. These constructions are said
to have a variable syntactic controller and a variable syntactic pivot. And
some languages, such as Tagalog (see Section 5.3.2), have many construc-
tions allowing a range of semantic roles access to the PSA. These construc-
tions have a semantic controller with no neutralization. As we will see, in
Tagalog there is no neutralization we can call S, nor even neutralization of
a single grammatical category of Actor or Undergoer in terms of the
marking on the predicate, as different types of actor and undergoer (e.g.
with different degrees of intention, agentivity, transitivity, affectedness)
are marked differently on the predicate depending on the nature of the
action and the Topic of the clause.5 In other constructions (e.g. the Reflex-
ive Construction) there is a generalized Actor, and it is an invariable
semantic controller. For ease of discussion we will generalize across the
different types of actor, patient and location in Tagalog by glossing the
various relevant forms as ‘Actor-Topic’, ‘Patient-Topic’ or ‘Location-
Topic’, respectively.
5.3.2 Alignment
The particular PSAs in the conjoined clause coreference construction dis-
cussed here are found in English, but many other languages, even closely
related ones, do not manifest PSAs in conjoined clause coreference construc-
tions, and so the determination of the relevant argument of the second

clause would be left to inference unconstrained by the grammar, and under-

goers in active clauses could be left to inference as easily as A and S. It is also
the case that some languages manifest constructions for conjoined clause
coreference with particular PSAs, but the restricted neutralization is [S,U]
rather than [S,A].6 Dyirbal, a Pama-Nyungan language spoken in north-
eastern Australia (Dixon 1972, 1980), is a well-known example of a language
that has grammaticalized an [S,U] restricted neutralization for conjoined
clause coreference. That is, the arguments participating in the coreference
must be U and/or S, but A does not participate in the coreference unless it is
cast as an S in the Dyirbal antipassive construction. Consider the examples
in (2) (adapted from Dixon 1980: 462).
(2) a. balan guda buŋa-n baŋgul yara-ŋgu bura-n

3sgf.abs dog.abs descend-pst 3sgm.erg man-erg see-pst
The dogcontroller went downhill and the man saw (pivot).
b. bayi yara buŋa-n buralŋanyu bagun guda-gu
3sgm.abs man.abs descend-pst see:pst:antip 3sgm.abs dog-dat
The mancontroller went downhill and (pivot) saw the dog.
c. balan guda baŋgul yara-ŋgu bura-n buŋa-n.
3sgf.abs dog.abs 3sgm.erg man-erg see-pst descend-pst
The man saw the dogcontroller and (pivot) went downhill.
These examples parallel the English examples in (1), but the interpretation
of the implied argument in the second clause of (2a) is obligatorily corefer-
ential with the Undergoer in the first clause, that is, it must be that the dog
is the controller and pivot (i.e. is the one that went downhill). In order to
have coreference that involves an A argument, the Dyirbal antipassive
construction must be used. This construction is an intransitive construction
with the verb marked with the antipassive marker ŋa, and has the Actor as
the single direct argument in the absolutive case and the Undergoer in the
dative case. Because Dyirbal has this antipassive construction, there is vari-
able access to the controller and pivot positions, and so the controller is a
variable syntactic controller and the PSA is a variable syntactic PSA, though
they manifest a different restricted neutralization from the corresponding
English construction. These grammaticalized constraints on interpretation
we have been looking at force a particular interpretation of an utterance in
both English and Dyirbal, but as the restricted neutralizations are different,
the interpretations are different. For example, if The man saw the dog and went
downhill is said in English, the interpretation has to be that the man went
downhill; but if the corresponding structure is used in Dyirbal, as in (2c), the
meaning has to be that the dog went downhill. We see here that the
construction must be taken as a whole, as it is the total construction
that influences the interpretation, and is not simply the sum of the individ-
ual words.
Where there is a choice of argument for PSA, the RRG theory of PSA
selection posits a default choice, depending on the Privileged Syntactic

Argument Hierarchy (3) and the privileged syntactic argument selection

principles (4): actor for PSA in so-called nominative-accusative alignment,
and Undergoer in ergative-absolutive alignment.
(3) Privileged Syntactic Argument Selection Hierarchy

arg. of DO > 1st arg. of do′ > 1st arg. of pred′ (x, y) > 2nd arg. of pred′ (x, y)
> arg. of pred′ (x)
(4) Privileged syntactic argument selection principles

a. Syntactically accusative constructions: highest-ranking macrorole is
default choice.
b. Syntactically ergative constructions: lowest-ranking macrorole is
default choice.
Yet it isn’t the case that all languages necessarily have such a default
choice. As Foley and Van Valin argued (1984, §4.3) Tagalog has many differ-
ent constructions for having different semantic roles as PSA, yet none is a
default choice. In all these languages where we have seen a choice of PSA,
the choice of pivot is determined by pragmatic factors, such as the identifia-
bility or topicality of the referent involved (the default is used when there is
no difference in terms of the pragmatic factors), and so we refer to these as
pragmatic PSAs, as opposed to those based strictly on semantic factors,
which we call semantic PSAs.
While the conjoined cross-clause coreference construction in English
manifests an [S,A] restricted neutralization, it is not the case that all con-
structions in English manifest a restricted neutralization, and it is not the
case that all of the constructions that do manifest a PSA in English manifest
the same restricted neutralization. For example, in the following example of
the English purposive construction the controller and pivot of the
construction are the Undergoers of the two clauses, not the Actor or S of
the clauses:
(5) He left this formcontroller for you to sign (pivot).
As in the construction in (1), the controller determines the reference of the

implied argument in the second clause, but in this case the referent referred
to appears as the traditional direct object (the Undergoer in RRG) of the first
clause, and is understood as the traditional direct object (the Undergoer in
RRG) of the second clause as well. This shows that the concept of PSA is not
the same as the traditional concept of ‘subject’.
Aside from the possibility of different restricted neutralizations, there is
also the possibility of unrestricted neutralization. Some languages, such as
Riau Indonesian (Gil 1994) and Mandarin Chinese (LaPolla 1993), have not
conventionalized any constraints on referent role identification of the type
associated with grammatical relations in any constructions in the lan-
guage (though they may have conventionalized other types of constraints
on interpretation). This means that in all the constructions we have looked

at interpretation is based entirely on inference from context and is not

constrained by the structure. In other languages, only some constructions
will have unrestricted neutralization. It is entirely language-specific and
construction-specific. In English, for example, the relative clause construc-
tion manifests neutralization, but no restriction, that is, any argument of
the modifying clause can appear as the head of this construction, as shown
in (6):
(6) a. the girl who[A] sang the song

b. the girl who[U] the police saved
c. the girl who[S] just came in
d. the girl to whom[non-macrorole core argument] the award was given
e. the car in which[peripheral argument: location] the man was held up
f. the car out of which[peripheral argument: source] the radio was stolen
In Tagalog, a Malayo-Polynesian language spoken in the Philippines, on

the other hand, there is a restricted neutralization for the Tagalog relative
clause construction: the head of the construction must be the grammatical
Topic of the modifying clause, and so it is a syntactic pivot for that construc-
tion. To explain this we will first give some background on the structure of
the Tagalog main clause.
Tagalog is a consistently focus-initial language and so in general the
predicate appears in initial position and the Topic appears (when it is not
a pronoun and not in focus) at the end of the clause, and the information
structure is Comment-Topic rather than Topic-Comment. The predicate in
most cases marks aspect, realis/irrealis, and the semantic role of the Topic of
the clause, and so the Topic is the controller of the semantic role marking (it
is a semantic and not syntactic controller because the marking differs with
each semantic role, i.e. there is semantic restriction but no neutralization
relative to each type of marking). Topic here is actually a grammatical
status, as it is an argument singled out for special morphological treatment,
as well as a pragmatic status, as it is what the clause is about. Generally,
almost any referent associated with the situation in some way, whether core
or peripheral argument semantically or even a very indirectly affected
referent, can be the Topic of the clause, though usually (but not always) it
is one that is identifiable to the hearer, and in many cases the predicate
takes a form to reflect the semantic role of that referent, and the marking of
the other references in the clause is usually also different.7 These different
constructions are not passive or antipassive, but simply different ways of
profiling the event (Foley and Van Valin 1984, §4.3), similar to the choice of
the A Construction vs. the O Construction in Jawarawa (Dixon 2000, 2004),
depending on what is considered to be what the clause or discourse is about,
but with more choices for Topic in Tagalog.
The representation of the Topic argument (if it appears in the clause and is
not a pronoun) generally takes either a marker of specificity (ang or si with
proper names of single referents) or a topic form of demonstrative pronoun

(most commonly ’yung) to mark it as the Topic. The set of pronouns, which
appear as second-position clitics, also distinguishes Topical from non-
Topical referents, with the latter appearing as possessive forms or dative
forms. In (7) there is a short natural conversation to exemplify this feature
(from my own fieldwork; see also Schachter (2008: 337–338) for sets of
constructed parallel examples with the same arguments but with different
choices of topic, and Schachter and Otanes (1972, Ch. 5) for many of the
major constructions used for marking different semantic roles).8
(7) 1. Jirehel: Madali lang kasi’ng gumawa ng salsa eh.

ma-dali lang kasi ang g<um>awa ng salsa eh
stat-easy just because spec <at>make poss sauce emphatic
‘Making sauce is easy.’
2. Wendy: Oo, madali lang.
oo ma-dali lang
yes stat-easy just
‘Yes, it’s easy.
3. Gawin mo lang ketsap,
gawa-in mo lang ketsap
do-pt:irr 2sg.poss just ketchup
You just make it with ketchup,
4. tapos lagyan mo ng tomatoes,
tapos lagay-an mo ng tomatoes
finish put-lt 2sg.poss poss tomatoes
then you add tomatoes,
5. lagyan mo ng salt and pepper to taste, tapos na.
lagay-an mo ng salt and pepper to taste tapos na
put-lt 2sg.poss poss salt and pepper to taste finish cs
add salt and pepper to taste; then, it’s done.’
In (7), the first speaker, Jirehel, refers to the making of sauce using a form
(gumawa) where the infix -um- marks it as Actor-Topic (it happens that in this
utterance the speaker has made the relevant clause the Topic of a higher
clause, but the phenomenon is the same), then the second speaker, Wendy,
uses the same root in line 3 of the example, but in the irrealis Patient-Topic
construction, to profile the event from the point of view with the sauce as
the Topic. She then follows this in lines 4 and 5 with two tokens of the root
lagay ‘put’ in the Location-Topic construction, to keep the sauce as the Topic,
but with the sauce now having the semantic role of the location where the
tomatoes and salt and pepper are to be added. Notice how there is no overt
reference to the Topic in any of Wendy’s utterances (e.g .’yung salsa ‘the
sauce’ could have been added to the end of each of Wendy’s utterances in
lines 3–5, but it wasn’t), yet we can tell what is being talked about (what is
the pragmatic and grammatical Topic) because of the marking on
the predicate.
We now can return to the issue of grammatical relations, that is,
restricted neutralizations. The choice of different roles as Tagalog Topic

affects the particular form of the construction and the behaviour of the
Topic in this and other constructions, such as the Tagalog relative clause
construction, where the head of the construction generally must be the
Topic of the modifying predicate (i.e. the form of the predicate must be
the same as if it were a main clause with that referent as Topic). For
example, if we were to recast the clauses in (7) as relative clause construc-
tions (and with realis predicates), we would get the forms in (8):
(8) a. lalaking gumawa ng salsa ‘man who made salsa’

lalake ¼ng g<um>awa ng salsa
man lnk <at.realis>make poss sauce
b. salsang ginawa ng ketsap ‘salsa made with ketchup’
salsa ¼ng g<in>awa ng ketsap
sauce lnk <pt.realis> poss ketchup
c. salsang nilagyan ng tomatoes ‘salsa in which tomatoes were put’
salsa ¼ng in-lagay-an ng tomatoes
sauce lnk realis-put-lt poss tomatoes
In Tagalog, then, for this construction there is a clear restricted neutral-

ization: the head of the construction must be the grammatical Topic of the
predicate that modifies it, unlike in English, where there is neutralization
but no restriction on what argument can be the head of an English relative
clause construction.
We saw that in Tagalog the controller of the semantic role marking on the
predicate (the Topic) is a semantic controller. English agreement, on the other
hand, manifests a syntactic controller. That is, the agreement is not with a
particular semantic role like in Tagalog, but manifests a restricted neutraliza-
tion, where the agreement is with the Actor or S of the clause, regardless of its
semantic or pragmatic role. We can see this from the examples in (9).
(9) a. I am helping Bill with his homework [agreement with A]

b. Bill is being held up by the activities [agreement with Undergoer as
passive S]
c. Bill is smiling [agreement with S Actor]
d. Bill is falling into a trap [agreement with S Undergoer]
e. There are many people in the park [agreement with non-topical S]
f. There are people grilling meat there [agreement with non-topical A]
Acehnese (Durie 1985, 1987), an Austronesian language spoken in Indo-

nesia, manifests a system of pronominal agreement on the verb with Actor-
marking prefixes and Undergoer-marking suffixes on the verb. The marking
is restricted to Actor and Undergoer, respectively, and there is no neutral-
ization of Actor and Undergoer, even with intransitive constructions. See,
for example, the following (from Durie 1987: 366, 369):
(10) a. (Gopnyan) geu-mat lôn b. (Lôn) lôn-mat-geuh

(3sg) 3-hold 1sg (1sg) 1sg -hold-3
‘(S)he holds me.’ ‘I hold him/her.’

c. Geu-jak (gopnyan) d. Lôn rhët(-lôn).

3-go (3sg) 1sg fall(-1sg)
‘(S)he goes.’ ‘I fall.’
From these examples we can see that the prefixes are used for the Actors of
transitive constructions (10a, b) and the single arguments of intransitive
constructions where the action is voluntary and so the argument is an Actor
(so-called SA), as in (10c), and the suffixes are used for the Undergoers of
transitive constructions and the single argument of intransitive construc-
tions where the action is involuntary (an Undergoer, so-called SP or SO), as
in (10d).9
5.3.3 Referent Tracking

The different choices of grammatical Topic we saw in Tagalog have the same
function in discourse as the variable syntactic pivots in English and Dyirbal
in terms of allowing for topic chains where an (often unmentioned) referent
can remain the topic across clauses even if its semantic role changes, and
the structure helps constrain the inference of the identity and role of the
relevant referent, as in (7). This type of referent-tracking mechanism is
known as a ‘switch function’ (Foley and Van Valin 1984: 321–374; Van Valin
1987; Van Valin and LaPolla 1997, §6.4), as the structure identifies the
difference in function of the referent while maintaining it as topic. There
are differences among languages, though, not only in the type of restricted
neutralization, such as in English vs. Dyirbal, but also in terms of what
referents can be an unmarked topic of the clause: if we contrast English,
Dyirbal, Tagalog and Chinese, we can see that English and Dyirbal severely
limit what can be topic in the unmarked clause structure ([S,A] or [S,U]
respectively); Tagalog allows just about anything relevant to be topic, but
marks it morphosyntactically, thereby constraining the inference of the
referent and its role; Chinese does not manifest any restricted neutraliza-
tions (LaPolla 1993) and so there is no grammatical restriction on what can
be topic in the unmarked clause structure, though there is no marking of it
as topic other than initial word order, and no marking of its role or identity,
and so identification of the referent and its role is not aided by the structure
(see, for example, the discussion of (12) below). This forms a separate though
related typological cline, from most restricted to least restricted in terms of
access to topic, aside from the typological cline related to the degree to
which the structure constrains the interpretation of the role and identity of
the referent being tracked.
Yet another type of restricted neutralization with variable PSAs, known as
the ‘switch-reference’ pattern, is found in Barai, a language of Papua New
Guinea (Olsen 1978, 1981; Van Valin 2009, §4), and in Choctaw, a Muskogean
language of North America (Heath 1975, 1977), among others. In this type of
system, when clauses are coordinated, the PSAs of the individual clauses
(defined differently in each construction and in each language) can be

coreferential or not, and there is marking on the predicate to show this (i.e.
whether the PSA is the same as in the following clause or different from the
following clause), as in the following Choctaw examples (adapted from
Heath 1977: 212):
(11) a. (0i)-0-pi:sa-ča:, 0i-iya-h ‘Hei sees himj and hei goes’

(3A)-3P-see-same 3A-go-present
b. 0-0i-pi:sa-na:, 0i-iya-h ‘Hei sees himj and hej/k goes’
3A-3P-see-different 3A-go-present
As can be seen from these examples, even without overt arguments, the affix
on the predicate marking whether the PSA in the following clause is the
same or different from that of the marked clause constrains the interpret-
ation of who is doing what. The pivots in this sort of system are generally
invariable syntactic pivots.
We have seen above that there are different kinds of constructions lan-
guages can have for constraining the inference of who is doing what, if they
have any at all. They may have some constructions that have invariable
semantic pivots (i.e. restriction with no neutralization), such as the verbal
marking in Tagalog and Acehnese, or invariable syntactic pivots, such as
reference across conjoined clauses in Warlpiri and Enga, or variable syntac-
tic pivots, such as reference across clauses in Dyirbal and English, and the
relative clause construction in Tagalog. The restricted neutralization found
in a construction could treat [S,A] the same in opposition to [U], or it may
treat [S,U] the same in opposition to [A]. As this is a construction-specific
phenomenon, the same language may have different constructions with
different PSAs, as we saw with Tagalog (see also Van Valin 1981 on Jakaltek
and Van Valin and LaPolla 1997: 282ff. on Tzutujil), and constructions with
different PSAs can sometimes be combined into a single complex structure,
such as when a Tagalog reflexive construction (invariable semantic pivot)
and any type of Tagalog clause structure (variable syntactic pivot) are com-
bined to form a complex structure. And of course a language may not have
any constructions that manifest PSAs, such as Mandarin Chinese and Riau
Indonesian.
5.4 How Grammatical Relations Develop
Lexico-grammatical structure becomes ‘grammaticalized’ or ‘lexicalized’

(becomes what we think of as grammar or words) through repeated use of
particular patterns to constrain the hearer’s interpretation of the
speaker’s intention in a particular way until it becomes conventionalized
on the societal level and habitualized on the individual level (LaPolla
2015). Ontogenetically we start with no structure, including no grammat-
ical relations, and in each society different types of constructions will
emerge out of the interactions of the speakers, and so each language is

unique in terms of what sorts of structures have conventionalized in the

language. There are some languages, such as Riau Indonesian (Gil 1994)
and Mandarin Chinese (LaPolla 1993), which have not conventionalized
constraints on referent identification of the type associated with gram-
matical relations (though they may have conventionalized other types of
constraints on interpretation). What this means is that there are no
conventionalized associations which relate position in word order, the
marking on the nouns or verb, and so on with particular semantic roles,
and so the structures of the language do not force particular interpret-
ations of the role of referents mentioned in discourse. The addressee can
still infer a context of interpretation in which the utterance will make
sense, but this inferential process is relatively unconstrained compared to
a language that has constructions that force a particular interpretation
within this functional domain, and so the addressee will have to rely
more on the assumptions of real-world semantics (what makes sense
given common knowledge about the world) to make sense of the utter-
ance. This does not mean that the inference won’t be influenced by
conversational implicatures. There may be common conversational impli-
catures that can influence the interpretation. For example, as there is a
rather strong frequency correlation between topic and actor in Chinese
(and many other languages), there is a conversational implicature that an
animate topic (the referent referred to by the utterance-initial referring
expression) is the actor of the clause. It is simply a conversational impli-
cature because it can be cancelled by the semantics of the referents or the
requirements of the context of interpretation, such as in (12):
(12) Xuéshēng yı̌jı̄ng fā-le chéngjı̄

student(s) already distribute-pfv grade
‘The students were already given their grades.’
This expression could potentially be understood as either ‘The students were

already given their grades’ or ‘The students already gave out grades (to
someone else)’, but it was understood in the context in which it was uttered
as ‘The students were already given their grades’ because students normally
receive grades, not give them out, and it made more sense in the context.
What happens in one type of conventionalization of grammatical relations
is that a conversational implicature of ‘actor as topic’ appears so often in
discourse that it becomes a conventional implicature, and then becomes so
strongly conventionalized that speakers cannot accept any other interpret-
ation.10 Note that all conventionalization (grammaticalization) is of con-
structions, not individual words, and not globally in the language (Bybee
2003; Himmelmann 2004; Gisborne and Patten 2011), so the RRG assump-
tion that grammatical relations are construction-specific is much more in
line with what we know about grammaticalization and with the facts of
languages, and so is more empirically sound than a view that posits abstract
global grammatical relations in all languages.

The difference between conversational implicature, conventional implica-

ture, and obligatory marking forcing a particular interpretation is the
degree to which speakers are free to use or not use the particular form to
constrain the hearer’s inferential process, and also the degree to which the
form forces a particular interpretation. Old English did not constrain the
identification of the role of a referent with word order, though it did
constrain the interpretation of referent role using a complex system of case.
Even so, the frequency with which reference to actors preceded the verb in
topic position led to a conversational implicature that gradually
strengthened as the case-marking system weakened, until we ended up with
the current system of Modern English, where word order alone constrains
the interpretation of the role of the main referents, and what was originally
the primary means of constraining the interpretation of the role of the
referent (the case marking) is now non-existent or, in the case of pronouns,
is now secondary, often assigned by word order.
Conventionalized constraints on the interpretation of coreference across
clauses also develop in a similar way. Initially there is no syntactic con-
straint on cross-clause coreference, and so the interpretation of which
referring expressions (including zero) co-refer is completely dependent on
inference from real-world semantics. For instance, in the following example
from Rawang, a Tibeto-Burman language of northern Myanmar, any of the
three coreference patterns given in the three translations would be possible,
and which would be correct would depend on the addressee’s inference of
which is most likely to be the interpretation intended by the speaker given
the addressee’s assumptions about hitting and crying and what is known
about the people involved.11
(13)
e phūŋí dɯ́ s ̀ ŋ dip bɯ́ à nɯ̀ ŋɯ̄ a:ʔmı̀
e e e
e phūŋ-í dɯ́ -s ̀ ŋ
e e dip bɯ́ -à
e nɯ̀ ŋɯ̄ -ap-ı̀
Apung-agt Adeu-loc hit pfv-tr.pst ps cry-tmdys-intr.pst
(i) ‘Apung hit Adeu and (Apung) cried’ or
(ii) ‘Apung hit Adeu and (Adeu) cried’, or
(iii) ‘Apung hit Adeu and (someone else) cried’
Utterances of this type are somewhat rare, though; more often only one
possible actor is mentioned, as in an utterance like John finished eating and
left, and so the conversational implicature that the actor is the same in both
clauses (and it is only an implicature at first) can become strengthened to
the point that it becomes conventionalized as the only possible interpret-
ation, as in English, where a clause such as John put the rock next to the
chameleon and turned brown has to mean that John turned brown, even if it
makes no sense, unlike in a language where this coreference pattern has not
conventionalized (e.g. Chinese) and so it would more likely be interpreted as
meaning the chameleon turned brown.
Agreement or cross-referencing on the verb develops as an unstressed
pronoun is reinforced by a stressed pronoun or full noun phrase often

enough for the unstressed pronoun to become cliticized to the verb. We

clearly see this process in Angami Naga, a Tibeto-Burman language of North-
east India (Giridhar 1980: 32, 59): the verbal prefixes (1sg ā-, 2sg n̂ -, 3sg puô-)
are transparently derived from the free pronouns (1sg ā, 2sg nō, 3sg puô), and
can be used together with the free pronouns, as in (14a), or with a noun
phrase, as in (14b) (see LaPolla 1992a, 1994 for other examples from Tibeto-
Burman languages).
ˇ ˇ
(14) a. nō n̂ -dōvı̄ b. nhıcunyô puô-dōvı̄
2sg 2sg-clever boy 3sg-clever
‘You are clever.’ ‘(The) boy is clever.’
Relational marking on noun phrases often arises as marking of location,

such as a locational noun, is used to constrain the inference of the relation-
ship of some referent to the state of affairs being predicated to a locational
sense, such as source, and then gets extended to the marking of other sorts
of participants (e.g. agents) through predictable pathways (see the following
section for more discussion, and also LaPolla 2004). This can further conven-
tionalize into fully paradigmatic case marking.
5.5 Why Syntactic Relations Develop
We have seen that syntactic relations develop from a form that is repeated
over and over again in discourse to the point that it becomes conventional-
ized as an obligatory part of the language, and thereby forces a particular
interpretation where otherwise there would be two or more possible inter-
pretations. But why would speakers repeat a form so often that this would
happen?
The answer lies in the culture of the speakers of the language, their way of
thinking, their value system. For a form to be used often enough for it to
become conventionalized, it must constrain the interpretation process of
the addressee in a way that is important to the speaker, so important that
the speaker is willing to put extra effort into constraining the addressee’s
inferential process in that particular way to make it more likely the
addressee will ‘get it right’. That is, the speaker wants to make sure the
addressee will infer that part of the communicative intention correctly,
more so, possibly, than other parts of the intention, and often uses a
particular form that they have used successfully before (and/or other people
have used successfully before) to constrain the interpretation in the same
way, over and over again. (We are creatures of habit and imitation, and
although we sometimes innovate, we more often go along with our usual
habits and also will imitate others.)
In the case of syntactic relations, what must be important to the speakers
is that the addressee correctly infer the roles of the major participants. The
clearest example of this is the development of relation morphology on the

noun phrase of the type agentive, patient, and/or anti-agentive.12 Marking of

participant role is, at least initially, marking of semantic role. In many of
the languages I’ve looked at (the Tibeto-Burman languages; LaPolla 1995,
2004), there is a clear development of agentive marking through the exten-
sion of ablative or instrumental marking to constrain the inference of which
participant is the agent. This begins only in contexts where there could be
confusion, such as when there are two human referents mentioned in an
utterance, and it is optional at that stage. The first speaker to do this would
have had the desire to constrain the interpretation of the semantic roles,
and in order to do so used a form already in the language (e.g. ablative
marking; it is easier to use material already in the language than to create
totally new material). Over time, this marking can become obligatory and
can also be extended to other sorts of agent-like referents. The motivation
for patient or anti-agentive marking is the same, but in the case of these
markers, the speakers chose to constrain the interpretation of the role of a
non-agent rather than an agent. In some of the older systems this type of
marking has developed beyond simple semantic marking, as speakers have
used material already in the language (the semantic marking) to constrain
the interpretation in new ways.
In some cases the pattern that gets conventionalized might not specifically
involve extra effort on the part of the speaker but simply reflects the dis-
course habits of the speakers (which again will reflect the culture of the
speakers). For example, in a culture where actors are very often made the
topic of conversation, and topics are mentioned in clause-initial position (also
a choice that influences the construction of the context of interpretation –
see LaPolla 2019), we might see this result over time in the conventionaliza-
tion of a word order constraint such as that in English. We can see this
tendency developing in some Tibeto-Burman languages, such as Qiang
(LaPolla with Huang 2003), but it has not yet fully conventionalized. For
example, in a Qiang transitive clause construction with two unmarked noun
phrases referring to human referents, usually the first one will be understood
as referring to the Agent, but pragmatics still controls word order more than
semantics, and so if some other referent is more topical than the Agent, the
noun phrase referring to the Agent will not appear first. In this kind of
situation the Qiang Agentive marking is often used to constrain the address-
ee’s interpretation of the relative roles. Agent-first is then the default and
unmarked situation, and could develop into an obligatory interpretation
with more reinforcement through repeated occurrence.
The motivation for the development of constraints on a particular func-
tional domain may not originally be part of the native culture but can
come though language contact: when people are bilingual in another
language that obligatorily constrains the interpretation of some func-
tional domain, such as the marking of source of information (i.e. has
obligatory evidential marking), and they use that language often enough
for the habit of constraining the evidential sense to become established,

they may eventually feel the necessity to constrain the interpretation of

source of information when using their own language. They can then
borrow the forms of the language that already has it, or use native material
for that purpose, and it may then develop into an obligatory category in
their own language. This is still repeated action based on the desire to
constrain the interpretation in a particular way leading to conventionali-
zation, but in this case the motivation came into the culture of the
speakers through the influence (habits) of another culture (see LaPolla
2009). Relevant to grammatical relations, the development of person
marking on the verb in some Tibeto-Burman languages seems to be related
to language contact (see LaPolla 2001).
Although all conventionalization has its origin in repeated actions that
have a cultural motivation, it is not always possible to find a direct link
between some motivation and the linguistic form post facto, especially if
the conventionalization happened in the far-distant past (though see the
papers in Enfield 2002 and De Busser and LaPolla 2015) and if there has been
considerable phonetic reduction of the forms used in the constructions. We
continue to use forms that are no longer transparently motivated just
because they are there, and are part of our habits of language use. We can
see this in the layering of marking, for example the fossilization and
maintenance of the -r plural in children, even though it is not seen as a plural
marker by most modern English speakers.13 The motivations for many
words used in English today are opaque to modern English speakers, such
as why we say dial to make a phone call, but they use the forms anyway. In
some cases, sound changes can make what was once transparently motiv-
ated opaque. For example, the modern word for ‘crow’ in Mandarin Chinese
is wū, which is not transparent, but when we look at the way it would have
been pronounced when it was first used (reconstructed as *ʔa), we can see
that at that time it was motivated as onomatopoeia. In terms of syntax, we
may have less evidence for the motivation of a particular word-order pat-
tern, but in some cases we can see the effect it has and possibly assume that
that effect was the motivation. For example, the English pattern of marking
mood with word order, that is, putting a different element in initial (Theme)
position in different moods,14 may have developed because of a desire to
clearly constrain the interpretation of mood.
References
Bloomfield, Leonard. 1914. An Introduction to the Study of Language. New York:

Henry Holt. (Photostatic reprint 1983 by John Benjamins, Amsterdam,
with Forward by Konrad Koerner and Introduction by Joseph Kess.)
Bybee, Joan. 2003. Mechanisms of change in grammaticalization: The role of
frequency. In Brian D. Joseph and Richard D. Janda (eds.), The Handbook of
Historical Linguistics, 602–623. Oxford: Blackwell.

Comrie, Bernard. 1981. Language Universals and Linguistic Typology. Oxford:

Blackwell.
Croft, William. 2001. Radical Construction Grammar: Syntactic Theory in Typo-
logical Perspective. Oxford: Oxford University Press.
Croft, William. 2013. Radical Construction Grammar. In G. Trousdale and T.
Hoffmann (eds.), The Oxford Handbook of Construction Grammar, 211–232.
De Busser, Rik and Randy J. LaPolla (eds.). 2015. Language Structure and Environ-
ment: Social, Cultural, and Natural Factors. Amsterdam: John Benjamins.
Dixon, R. M. W. 1972. The Dyirbal Language of North Queensland. Cambridge:
Dixon, R. M. W. 1980. The Languages of Australia. Cambridge: Cambridge
University Press.
Dixon, R. M. W. 2000. A-constructions and O-constructions in Jarawara.
International Journal of American Linguistics 66(1): 22–56.
Dixon, R. M. W. 2004. The Jarawara Language of Southern Amazonia. Oxford:
Dryer, Matthew. 1997. Are grammatical relations universal? In Joan Bybee,
John Haiman and Sandra A. Thompson (eds.), Essays on Language Function
and Language Type, 115–143. Amsterdam: John Benjamins.
Durie, Mark. 1985. A Grammar of Acehnese. Dordrecht: Foris.
Durie, Mark. 1987. Grammatical relations in Acehnese. Studies in Language 11:
365–399.
Enfield, Nicholas. 2002. Ethnosyntax: Explorations in Grammar and Culture.
Foley, William A. and Robert D. Van Valin, Jr. 1977. On the viability of the
notion of ‘subject’ in universal grammar. Proceedings of the Berkeley Linguis-
tics Society 3: 293–320.
Gabelentz, H. Georg C. von der. 1869. Ideen zu einer vergleichenden
Syntax. Wort- und Satzstellung. Zeitschrift für Völkerpsychologie und Sprach-
wissenschaft 6: 376–384.
Gary, Judith Olmstedt and Edward L. Keenan. 1977. On collapsing grammat-
ical relations in universal grammar. In P. Cole and J. M. Sadock (eds.),
Grammatical Relations (Syntax and Semantics 8), 83–120. New York: Aca-
demic Press.
Gil, David. 1994. The structure of Riau Indonesian. Nordic Journal of Linguistics
17: 179–200.
Giridhar, P. P. 1980. Angami Grammar (CIIL Grammar Series 6). Mysore:
Central Institute of Indian Languages.
Gisborne, Nikolas and Amanda Patten. 2011. Construction Grammar and
grammaticalization. In Heiko Narrog and Bernd Heine (eds.), The Oxford
Handbook of Grammaticalization, 92–104. Oxford: Oxford University
Press.

Halliday, M. A. K. 1994. An Introduction to Functional Grammar (2nd ed.).

London: Arnold.
Heath, Jeffrey. 1975. Functional relationships in grammar. Language 51(1):
89–104.
Heath, Jeffrey. 1977. Choctaw cases. Proceedings of the Berkeley Linguistics Society
3: 204–213.
Himmelmann, Nikolaus P. 2004. Lexicalization and grammaticization:
Opposite or orthogonal? In Walter Bisang, Nikolaus P. Himmelmann
and Björn Wiemer (eds.), What Makes Grammaticalization? A Look from Its
Components and Its Fringes, 21–42. Berlin: Mouton de Gruyter.
Hopper, Paul. 1991. On some principles of grammaticization. In Elizabeth C.
Traugott and Bernd Heine (eds.), Approaches to Grammaticalization, Vol. 1,
17–36. Amsterdam: John Benjamins.
Jespersen, Otto. 1909–1949. A Modern English Grammar on Historical Principles,
Vols. I–VII. London: Allen & Unwin.
Keenan, Edward L. 1976. Towards a universal definition of ‘subject’. In
Charles N. Li (ed.), Subject and Topic, 305–333. London: Academic Press.
LaPolla, Randy J. 1992a. On the dating and nature of verb agreement in
Tibeto-Burman. Bulletin of the School of Oriental and African Studies 55(2):
298–315.
LaPolla, Randy J. 1992b. Anti-ergative marking in Tibeto-Burman. Linguistics
of the Tibeto-Burman Area 15(1): 1–9.
LaPolla, Randy J. 1993. Arguments against ‘subject’ and ‘direct object’ as
viable concepts in Chinese. Bulletin of the Institute of History and Philology 63
(4): 759–813.
LaPolla, Randy J. 1994. Parallel grammaticalizations in Tibeto-Burman: Evi-
dence of Sapir’s ‘drift’. Linguistics of the Tibeto-Burman Area 17(1): 61–80.
LaPolla, Randy J. 1995. On the utility of the concepts of markedness and
prototypes in understanding the development of morphological systems.
Bulletin of the Institute of History and Philology 66(4): 1149–1185.
LaPolla, Randy J. 2001. The role of migration and language contact in the
development of the Sino-Tibetan language family. In R. M. W. Dixon and A.
Y. Aikhenvald (eds.), Areal Diffusion and Genetic Inheritance: Case Studies in
Language Change, 225–254. Oxford: Oxford University Press.
LaPolla, Randy J. 2004. On nominal relational morphology in Tibeto-Burman.
In Ying-jin Lin, Fang-min Hsu, Chun-chih Lee, Jackson T.-S. Sun, Hsiu-fang
Yang and Dah-an Ho (eds.), Studies on Sino-Tibetan languages: Papers in Honor
of Professor Hwang-cherng Gong on his seventieth birthday, 43–74. Taipei: Insti-
tute of Linguistics, Academia Sinica.
LaPolla, Randy J. 2006. On grammatical relations as constraints on referent
identification. In Tasaku Tsunoda and Taro Kageyama (eds.), Voice and
Grammatical Relations: Festschrift for Masayoshi Shibatani (Typological Studies
in Language), 139–151. Amsterdam: John Benjamins.
LaPolla, Randy J. 2009. Causes and effects of substratum, superstratum and
adstratum influence, with reference to Tibeto-Burman languages. In

Yasuhiko Nagano (ed.), Issues in Tibeto-Burman Historical Linguistics (Senri

Ethnological Studies 75), 227–237. Osaka: National Museum of Ethnology.
LaPolla, Randy J. 2010. Hierarchical person marking in Rawang. In Dai
Zhaoming (ed.), Forty Years of Sino-Tibetan Language Studies: Proceedings of
ICSTLL-40, 107–113. Harbin, China: Heilongjiang University Press.
LaPolla, Randy J. 2015. On the logical necessity of a cultural connection for
all aspects of linguistic structure. In Rik De Busser and Randy J. LaPolla
(eds.), Language Structure and Environment: Social, Cultural, and Natural Factors,
LaPolla, Randy J. 2019. Arguments for seeing Theme-Rheme and Topic-
Comment as separate functional structures. In J. R. Martin, Y. Doran and
G. Figueredo (eds.), Systemic Functional Language Description: Making Meaning
Matter. London: Routledge.
LaPolla, Randy J., with Chenglong Huang. 2003. A Grammar of Qiang, with
Annotated Texts and Glossary (Mouton Grammar Library 39). Berlin: Mouton de
Gruyter.
Marantz, Alec P. 1982. Grammatical relations and explanation in linguistics.
In Annie Zaenen (ed.), Subjects and Other Subjects: Proceedings of the Harvard
Conference on the Representation of Grammatical Relations, 1–24. Bloomington,
IN: IULC.
Marantz, Alec P. 1984. On the Nature of Grammatical Relations (Linguistic Inquiry
Monograph 10). Cambridge, MA: MIT Press.
Platt, J. F. 1971. Grammatical Form and Grammatical Meaning: A Tagmemic View
of Fillmore’s Deep Structure Concepts. Amsterdam: North-Holland.
Olson, Mike. 1978. Switch-reference in Barai. Proceedings of the Berkeley Linguis-
tics Society 4: 140–157.
Olson, Mike. 1981. Barai Clause Junctures: Toward a Functional Theory of Inter-
clausal Relations. PhD dissertation, Australian National University.
Schachter, Paul, revised by Lawrence A. Reid. 2008. Tagalog. In Bernard
Comrie (ed.), The World’s Major Languages (2nd ed.), 833–855. London:
Routledge.
Schachter, Paul and Fei T. Otanes. 1972. A Tagalog Reference Grammar. Berke-
ley, CA: University of California Press.
Seuren, P. A. M. 1998. Western Linguistics: An Historical Introduction. Oxford:
Blackwell.
Van Valin, Robert D., Jr. 1977. Ergativity and the universality of subjects.
Proceedings of the Chicago Linguistic Society 13: 689–706.
Studies in Language 5(3): 361–394.
Van Valin, Robert D., Jr. 1987. Aspects of the interaction of syntax and
pragmatics: Discourse coreference mechanisms and the typology of gram-
matical systems. In Jef Verschueren and Marcella Bertuccelli-Papi (eds.),
The Pragmatic Perspective: Selected Papers from the 1985 International Pragmatics
Conference (Pragmatics and Beyond Companion Series 5), 513–531. Amsterdam:
John Benjamins.

controllers. In Lilián Guerrero Valenzuela, Sergio Ibáñez and Valeria
A. Belloro, (eds.), Studies in Role and Reference Grammar, 45–68. Mexico:
UNAM.
Notes
1 The name ‘Role and Reference Grammar’ derives from the early focus on
the semantic roles and pragmatic functions in discourse referent tracking
of grammatical relations.
2 Following best practice in typology, for language-specific (descriptive)
categories and constructions I will capitalize the initial letters of the
name of the category or construction, but for comparative concepts
I will not capitalize the first letter. So, for example ‘Actor’ refers to
the language-specific grammatical category manifesting a particular
neutralization of semantic roles in the language under discussion,
while ‘actor’ refers to the comparative concept of the one who per-
forms an action. As there are no universal or cross-linguistic gram-
matical categories, descriptive and comparative concepts need to be
kept distinct.
3 Note that the identification of the referent of it as the same as that of the
newspaper is not due to grammatical relations, but simply to inference;
there is nothing in the grammar that obligatorily constrains the inter-
pretation, the way the inference of the relationship between Jim and the
thrower of the newspaper is constrained by the grammar.
4 English S is itself a neutralization of semantic macroroles for grammat-
ical purposes that is not found in all languages (see below on Acehnese
and Tagalog). Actor and Undergoer are also language-specific restricted
neutralizations of semantic roles for grammatical purposes, hence are
called macroroles, but are at a lower level than the restricted neutraliza-
tions of macroroles we are talking about here. In English and many other
languages there is also variable access to Undergoer status when there is
both a theme and a recipient or location in the clause, e.g. in the con-
struction I gave the book to Mary, the book is the Undergoer, but in the
construction I gave Mary the book, Mary is the Undergoer, and in Load the
truck with hay, the truck is the Undergoer, while in the construction Load the
hay on the truck, the hay is the Undergoer. The choice of one construction or
the other usually depends on the relative topicality of the referents. See
Van Valin and LaPolla 1997: 144ff.

5 For example, in a Tagalog single argument clause, semantically different

arguments can be marked differently, e.g. given the root dulas, madulas
can be used for ‘slip (unintentionally)’ and dumulas can be used for ‘slide
(i.e. slip intentionally)’, and the marking of an intransitive actor can be
different from that of a transitive actor, even with the same root, e.g.
labas ‘outside’: lumabas ‘come/go out’ vs. maglabas ‘bring/take out’.
6 In the typology literature the [S,A] restricted neutralization is often
referred to as ‘nominative-accusative alignment’, as in Latin the
A and S take the nominative case in opposition to the U, which takes
the accusative case, and the [S,U] restricted neutralization is often
referred to as ‘ergative-absolutive alignment’, as in Dyirbal the U and
S take the absolutive case in opposition to the A, which takes the
ergative case. In some of the literature, languages are talked about as
having one of these alignments, but a more empirical approach is to
look at the individual constructions, as not all constructions in the
language necessarily manifest the relevant alignment, and a single
language can have different constructions manifesting different align-
ments (see Van Valin 1981 on Jakaltek and Van Valin and LaPolla 1997:
282ff. on Tzutujil).
7 An example of an indirectly affected referent as Topic would be Huwag
mong ubusan ng gasolina si Ricky [negimp 2sgnt¼lnk finish-lt poss
gasoline spec pn] ‘Don’t use up all the gasoline on Ricky’ (using the
Location Topic construction – cf. the use of the locative expression ‘on X’
in the English translation for the one affected).
8 Abbreviations used in the Tagalog examples: at Actor-Topic infix or
prefix; cs Change of State marker; irr Irrealis; lnk clitic Linker; loc
Locative marker; lt Locative-Topic suffix; poss Possessive linker; pt
Patient-Topic suffix or infix or prefix; redup Reduplication of initial
syllable for marking imperfective and planned actions; spec Specific
referent; stat Stative predicate.
9 In some languages there is another type of person marking on the
predicate that is not based on semantic role or PSA, as the marking
reflects the speaker, and possibly hearer and third-person argument
based on a person hierarchy such as 1 < 2 < 3. The predicate may also
have marking for when the actor is lower on the hierarchy than the
undergoer. This type of system is called hierarchical marking. See
LaPolla (2010) for one example.
10 Even as a conversational implicature the default interpretation can be
very strong. For example, the implicature that actions occur in the order
that they are talked about is quite strong in English, and so the average
speaker would say that they got married and had a baby means something
different from they had a baby and got married, but the implicature can be
cancelled, e.g. by adding but not in that order after either of the two
possible orders.

11 Abbreviations used in the Rawang example: agt agentive marker; intr.

past intransitive past-tense marker; loc locative marker; pfv perfective
aspect marker; ps predicate sequence marker (marks non-final clause);
tmdys time marker (marks a past action as having occurred within the
past few days); tr.past transitive past-tense marker. In the Rawang
example in (13) the tones are high á, mid ā, low à. All syllables that
end in a stop consonant are in the high tone. Open syllables without a
tone mark are unstressed. A colon marks non-basic long vowels. In the
Angami examples in (14), the tones are mid-level ā and low falling â.
12 Anti-agentive marking differs from patient marking in that it is not
marking what role a particular referent has, but what role it does not
have: it marks the mention of a human referent (at least human patients
and datives, but sometimes possessors as well) as not being agents. See
LaPolla (1992b, 2004).
13 George W. Bush famously said ‘Childrens do learn when standards are
high and results are measured’ (Reuters, 26 Sep 2007), showing that even
the -en plural does not constrain the sense of plurality enough for him.
This is an example of layering (Hopper 1991: 22), and we can see that
layering occurs when someone feels the existing marking is not con-
straining the inference of that particular semantic domain enough.
14 In unmarked cases, in declarative mood the subject is in initial position;
in polarity interrogatives, the polarity-marked auxiliary occurs in initial
position; in question-word questions, the question word appears in ini-
tial position; in imperatives the verb appears in initial position. So the
addressee’s inference of the mood of the clause (which generally repre-
sents how the speaker is interacting with the addressee) is greatly con-
strained by the first constituent (the Theme), and this influences the
interpretation of the rest of the utterance. Theme (as in Theme-Rheme) is
important precisely because the inferences drawn after hearing the
Theme influence the rest of the interpretation. (See Halliday 1994,
Ch. 3; LaPolla 2019.)

6
Argument Structure
Alternations
James K. Watters
Abbreviations

ANTIC anticausative MOOD mood

ASP aspect marker (general) NONFUT non-future
CON continuative PREP preposition
CONN connective PRON pronominal
DIR direction, secondary theme PV preverb (Abaza)
ELAT elative SP subject prefix (Bantu)
6.1 Introduction
In the most general sense, ‘argument structure’ refers to ‘the specification

of and relation between a word’s semantic and syntactic arguments’ (Jack-
endoff 2002: 134). Specifying the arguments that are semantically deter-
mined by a predicate and determining the relationship between those
arguments and their syntactic realization has been a key field of research
in all current linguistic theories. We can imagine a spectrum of possibil-
ities, from the possibility that all the details of a word’s syntax follow from
a correct account of its semantics to the other extreme claiming that the
relation between the semantic and syntactic configurations are simply
arbitrary. Role and Reference Grammar (RRG) takes a middle course in
accounting for the relationship between a verb’s meaning and its syntax:
the syntactic argument structure is projected from the verb’s lexical infor-
mation, along with other predicates in the clause, and is also modified by
discourse pragmatics.

Argument Structure Alternations 293
The purpose of this chapter is to give an introductory account of how

verbal argument structures and their alternations are handled in RRG.* The
literature across linguistic theories on these issues is vast and there is no
space to review other models or make detailed comparisons or to discuss
argument structures of non-verbal predicates.
RRG proposes a logical structure (LS) associated with each verb that is
based on a system of lexical decomposition. Unlike the lexical decompos-
ition found in Generative Semantics, or, more recently in Distributed
Morphology, the decomposition is largely defined by Aktionsart or lexical
aspect (Vendler 1967[1957]). Following work by Dowty (1979), RRG applies
semantic tests to determine a verb’s syntactically relevant lexical
decomposition.
The alternations in the syntactic arrangements of a verb’s arguments
complicate attempts to generalize about the verb’s argument structure. In
the Chomskyan tradition, such alternations are often accounted for by a
derivational process involving movement. RRG, however, is a monostratal
theory in which each sentence has a single morphosyntactic representation
linked by a set of rules to its semantic representation. It is important to note
that the mapping between the semantic and syntactic representation does
not involve movement or syntactic derivation of any kind.
Argument structure alternations involve different syntactic alignments
among arguments projected by the LS and may also involve adjuncts (the
elements of the peripheries of the layered structure of the clause). The order
of presentation in this chapter is based in part on the distinction between
lexical and syntactic aspects of argument structure. After some introductory
comments in Section 6.2, Section 6.3 focuses on lexical processes that affect
the linking of arguments to macroroles. Section 6.4 looks at syntactic pro-
cesses that affect the linking of arguments or adjuncts to the privileged
syntactic argument (PSA). Section 6.5 discusses two processes, the passive
and antipassive, that reflect marked mappings at both levels: they can have a
lexical effect on the linking to macroroles and also a syntactic effect,
determining which argument functions as the PSA.
The examples in this chapter come from a selection of languages from
different linguistic families, including some that have been discussed before
in the RRG literature. However, most of my fieldwork has been in the
Tepehua branch of the Totonac-Tepehua language family (Mexico). Because
of this, apart from English examples, illustrations of some of the more
complex issues will often be drawn from examples in Tepehua.1
A key concept in RRG is the Completeness Constraint, which places an
important restriction on argument structure alternations.
* This chapter has benefited significantly from comments by a peer reviewer and suggestions by Delia Bentley. Of course, any
remaining mistakes and obscurities are my own fault.

294 JAMES K. WATTERS
(1) Completeness Constraint:

(Van Valin 2005: 129–130)
This constraint allows for a bidirectional mapping between the syntactic

structure of the clause and the lexical structure. Two points of clarification:
(a) Van Valin points out that ‘explicitly specified’ means that a variable or a
constant fills the argument position in the logical structure; if it is filled by
‘Ø’, it is unspecified (2005: 130); and (b) note that only referring expressions
in the syntax need to be linked to an argument in the semantic representa-
tion. These points will be relevant in some of the discussion in this chapter.
6.2 Argument Structure Alternations
Consider the verb give in English or its equivalent in some other language. It
evokes a scene that includes a giver, something given, and a person or object
that receives what is given. As part of the meaning of the verb, these
arguments are included in the lexical entry for give but the arguments
may be realized in different ways syntactically. For example, argument
structure alternations discussed in the literature on English syntax include
examples such as the following.
(2) a. Randy gave the book to Kim.

b. Randy gave Kim the book.
This particular kind of alternation, commonly known as the dative alternation,

is a lexical alternation, affecting the mapping from the LS to the undergoer. As
will be seen in 6.3.2, not all languages have such an alternation.
Other alternations in the argument structure of give in English are
possible.
(3) a. The book was given to Kim (by Randy).

b. Kim was given the book (by Randy).
c. Randy gave the book.
d. Randy was always giving.
In each of these examples, the sentence entails that someone gave, some-
thing was given, and someone was the recipient or intended recipient (see
Williams 2015: 199–202). The argument structure is shaped by the passive
constructions in (3a) and (3b). The examples in (3c) and (3d) show that each
may have one unspecified argument but, unlike many languages, without
any morphology to signal the change. The syntactic arrangement of the verb
and its arguments is significantly different in each case.

The linking of arguments has been a key element of RRG since its incep-
tion (Foley and Van Valin 1984). As pointed out by Van Valin and LaPolla
(1997: 384, 389–392), the linking between semantic arguments and their
position in the syntax has two major phases:
1. mapping the arguments in logical structures onto macroroles, and

2. mapping the macroroles and other arguments onto the syntax.
The core arguments are those that are determined by the semantic decom-
position of the verb. Adjuncts are non-argument PPs and adverbs, which
occur in a periphery. RRG recognizes a third class of argument, the ‘argu-
ment-adjuncts’, which will be presented in 6.4.
The syntactically relevant semantics of a sentence is represented in the LS
(see Chapter 3 of this volume). That semantic representation determines the
kind of syntactic template (e.g. tree structure) that the sentence maps onto,
following a default principle.
Syntactic template selection principle:
The number of syntactic slots for arguments and argument-adjuncts within the
core is equal to the number of distinct specified argument positions in the
semantic representation of the core. (Van Valin 2007: 130)
There are typically language-specific restrictions, as well, such as the English

constraint that all cores have a syntactic valence of 1 (i.e. English requires
dummy subjects for predicates without any semantic argument, such as
rain).
6.3 Lexical Alternations
The standard mapping of a verb’s arguments onto the actor and undergoer
macroroles follow the Actor-Undergoer Hierarchy (AUH; Chapter 4). There
are, however, processes that are common cross-linguistically that alter the
standard argument structure or alignment. These include the two types of
constructions discussed in this section: noun incorporation and the ditran-
sitives or dative constructions.
6.3.1 Noun Incorporation

A common feature of head-marking languages is noun incorporation. In
some languages, such as West Greenlandic, noun incorporation is so pro-
ductive that it is reported ‘there are infinitely many possible forms involving
noun incorporation’ (Sadock 1991: 84).
We can start with an example from Nahuatl (Uto-Aztecan) given by Sapir
(1911: 260) in one of the first discussions of noun incorporation.

(4) a. ni-c-qua in nacatl

1sbj-3obj-eat the flesh
‘I eat the flesh.’
b. ni-nica-qua
1sbj-flesh-eat
‘I flesh-eat.’
In (4a) the verb is clearly transitive, requiring the direct object pronom-
inal form. The verb in (4b), however, is intransitive and so lacks the
direct object marking. Now consider the following examples from
Tongan (Polynesian) (5a–b) and Yucatec Maya (6a–b) (from Mithun 1984:
851, 857).
(5) a. Na'e inu 'a e kavá -'é Sione.

pst drink abs conn kava erg John
‘John drank the kava.’
b. Na'e inu kava 'a Sione
pst drink kava abs John
‘John kava-drank.’
(6) a. t-in-č’ak-Ø-ah če'.

comp-I-chop-it-prf tree
‘I chopped a tree.’
b. č'ak-če'-n-ah-en.
chop-tree-antip-prf-I(abs)
‘I wood-chopped.’
In both of these alternations, the verb is intransitive when the noun is

incorporated, as can be seen by the absolutive case for the actor in both
(5b) and (6b). (The intransitivity of the Yucatec Maya form in (6b) is also
marked by the antipassive suffix, which is typically present in the derivation
of an intransitive verb from a transitive.)
The key difference regarding semantic arguments in (4), (5) and (6) is that
the unincorporated patient is referential and maps onto the undergoer
position in a transitive construction. The incorporated nouns, however, are
non-referential and therefore do not correspond to any specific argument in
the LS, resulting in intransitive sentences. This kind of noun incorporation,
therefore, has a detransitivizing effect.
Another type of noun incorporation – perhaps the most common cross-
linguistically (McGregor 1997) – is the incorporation of nouns referring to
body parts. These constructions typically do not change a transitive verb
into an intransitive. Rather than deleting the undergoer macrorole, they
display an alternation in the linking to undergoer: in the non-
incorporated forms, the body part is the undergoer; in the incorporated
forms, the undergoer is the person, the possessor of the body part. In
some languages the only nouns that can be incorporated into the verb are
body parts.

Body part incorporation
results in part from the frequent recurrence and natural cohesion of

many activities affecting parts of the body, e.g. ‘to hand-wash’ or ‘to
tooth-brush’. In addition, noun incorporation of body parts allows
affected persons to assume a primary case role, such as subject or direct
object, rather than merely oblique possessor.
(Mithun 1984: 858)
The person affected can assume ‘a primary case role’ in these constructions
because the transitivity of the verb is unchanged, with the possessor of the
body part linking to the undergoer macrorole, as the ʻaffected personʼ.2
Mithun (1984: 857) provides the following example from Tupinamba (Tupí).
(7) a. s-oβá a-yos-éy

his-face I-it-wash
‘I washed his face.’
b. a-s-oβá-éy
I-him-face-wash
‘I face-washed him.’
Unlike the previous examples of noun incorporation, the verb continues

to be transitive – both forms in (7) have an actor and undergoer. Both
sentences have the same LS but differ in regard to undergoer selection.
(8) a. wash′ (a-, [have.as.part′ (s-, oβá)]) ¼ (7a)

b. wash′ (a-, [have.as.part′ (s-, oβá)]) ¼ (7b)
In both examples, the first person singular prefix a- maps onto the actor
macrorole. In (8a) ‘his face’, as patient, is the unmarked choice for under-
goer. The incorporated form in (8b) requires the marked choice for under-
goer and the possessor maps onto the undergoer macrorole.
One other kind of noun incorporation needs to be briefly discussed here.
At the beginning of this section, it was mentioned that Greenlandic Eskimo
is known to have extremely productive noun incorporation (Sadock 1991).
A language with similar constructions is Southern Tiwa (Kiowa-Tanoan) and
both languages are discussed in Rosen (1989). Consider the following
examples from Allen et al. (1984: 297).
(9) a. Wisi bi-musa-tuwi-ban

two 1sg:B-cat-buy-pst 3
‘I bought two cats.’
b. Yedi ibi-musa-tuwi-ban
those cats-B:B-buy-pst
‘They bought those cats.’
These examples, like many in Allen et al. (1984) and Sadock (1991), are
notably different from the noun-incorporation examples discussed previ-
ously, in at least two ways. First, the incorporated nouns are modified by a
quantifier (9a) and a deictic (9b), and, second, they are clearly referential.

Van Valin and LaPolla (1997: 66–68) present an analysis of similar construc-
tions in Greenlandic Eskimo, showing how RRG accommodates such struc-
tures with distinct operator and constituent projections in the layered
structure of the noun phrase. The significant conclusion is that, unlike the
earlier examples discussed in this section, these do not involve the incorpor-
ation of a bare noun, but rather full RPs, as shown by the fact that they are
referential and by the presence of modifiers.
6.3.2 Ditransitives
For our purposes here, I will assume the definition of ditransitives offered by
Malchukov et al. (2010: 1):
A ditransitive construction is defined here as a construction consisting

of a (ditransitive) verb, an agent argument (A), a recipient-like argument
(R), and a theme argument (T).
As they point out, this excludes some verbs of transfer which do not involve
a recipient (such as put) as well as benefactives (discussed in Section 6.4); but
it includes verbs with a ‘recipient’ in both a literal and an extended sense.
These are verbs that in English manifest the well-known dative alternation.
In RRG, ditransitives and other verbs with three arguments require a
complex representation with more than one predicate. The common LS for
a ditransitive involves two states of affairs joined by CAUSE.
(10) [do′ (x, Ø)] CAUSE [BECOME have′ (y, z)]
Following the AUH, the first argument of do′, x, is the actor, while the
theme, the second argument of have′, z, is the undergoer.
Over the last fifty years, there have been many studies of the English
dative alternation that try to account for the semantic similarities and
differences between pairs of sentences such as those in (2) and (11).
(11) a. Chris sent the money to Sam.

b. Chris sent Sam the money.
It should be noted that dative alternation is absent from many languages.

Consider the following examples from Alacatlazala Mixtec, an Otoman-
guean language of Mexico (from Zylstra 1991: 13) and from French.
(12) táshı̄ i tūtū ndāha sı̄hí ñá

(con)give I paper hand mother her
‘I’m giving the paper to her mother.’
(13) sı̄kō ñá noní noo i

(compl)sell she corn face my
‘She’s selling corn to me.’
(14) Jean a donné le livre à Marie

John has given the book to Mary
‘John gave the book to Mary.’

(15) *Jean a donné Marie le livre.

John has given Mary the book
‘John gave Mary the book.’
Unlike English, Mixtec and French do not have dative alternation. In

Mixtec languages the indirect object requires a body-part relational noun,
functioning as a preposition (p.c. Carol Zylstra). If the indirect object follows
the verb the result is ungrammatical. Likewise in French, a construction
similar to the English ‘dative movement’ is ungrammatical. In both lan-
guages, the direct object behaves like the direct object in a simple transitive
clause and the indirect object must be preceded by a preposition (or a
relational noun, in the case of Mixtec). These exemplify what Dryer (1986:
815) calls a ‘direct object’ language: the undergoer of the simple transitive
continues to manifest the same morphosyntactic properties in the ditransi-
tive construction (the ‘direct object’ is still the ‘direct object’). Such ‘direct
object’ languages require the unmarked undergoer choice predicted by the
AUH in ditransitives, choosing theme or patient as undergoer rather than
the recipient.
However, some languages require a marked undergoer choice in ditransi-
tives: the recipient, not the theme, is consistently linked to the undergoer
macrorole. Dryer (1986: 815) calls these ‘primary object languages’. Peterson
(2007: 144) provides the example of Hakha Lai, a Tibeto-Burman language.
(16) a. vok na-Ø-hmuʔ

pig 2sg.sbj-3sg.obj-see
‘You saw the pig.’
b. na-ka-hmuʔ
2sg.sbj-1sg.obj-see
‘You saw me.’
c. vok na-ka-peek
pig 2sg.sbj-1sg.obj-give
‘you gave me the pig.’
The verb agrees with the actor and the undergoer in the simple transitive
clauses (16a) and (16b) and with the recipient in the ditransitive clause (16c).
This agreement pattern in the ditransitive is obligatory – the alternation
found in English dative movement is absent. (For further discussion of
Tibeto-Burman examples, see Van Valin and LaPolla 1997: 272–273.) In
‘primary object languages’, unlike the Mixtec and French examples, the
recipient is obligatorily linked to the undergoer macrorole.
In Van Valin and LaPolla (1997: 387) this marked choice for undergoer is
attributed to animacy – the animate indirect object is chosen over the
inanimate direct object for undergoer position (see, however, Guerrero
and Van Valin 2004 for further discussion).
Some languages, such as Mixtec and Tepehua, have no syntactically
ditransitive verb roots. Although Tepehua has no verb roots with three
direct core arguments, the LS for predicates such as ma:laqatʃa: ‘send’ (17a)

and stʼa: ‘sell’ (18), like their counterparts in other languages, require three
semantic arguments. Consider the examples in (17) and (18).
(17) a. ma:laqatʃa:-ɬ
send-pfv
‘s/he sent it’
b. ma:laqatʃa:-ni-ɬ
send-dat-pfv
‘s/he sent it to him/her’
(18) a. ʃtaq-ɬi
give-pfv
‘s/he gave it.’
b. ʃtaq-ni-ɬ
give-dat-pfv
‘s/he gave it to him/her’
Totonac-Tepehua languages have only one (or two) adpositions, with

applicatives fulfilling a preposition-like function (Watters 2019). Rather than
employing a preposition to refer to the recipient, Totonac-Tepehua lan-
guages use the suffix -ni, signalling a marked undergoer choice. The appli-
cative -ni licenses a third argument.
In (17) and (18), both the transitive and the ditransitive forms have the
basic LS given in (10). The only difference is that the transitive constructions
in (17a) and (18a) do not specify the y-argument (the recipient) in the LS.
(19) a. [do′ (x, Ø)] CAUSE [BECOME have′ (Ø, z)] (¼ (17a), (18a))
b. [do′ (x, Ø)] CAUSE [BECOME have′ (y, z)] (¼ (17b), (18b))
Following the Completeness Constraint (1) the unspecified recipient in (19a)

is absent from the syntax; in (19b) the recipient is specified in the LS, so must
also be syntactically present. The recipient argument cannot appear in a
clause unless the applicative suffix, -ni, occurs on the verb.
In the case of the verbs in (17a) and (18a), the theme is linked to the
undergoer position and what would correspond to the indirect object is
unexpressed. In both cases the simple transitive forms of such verbs are
typically used in contexts where the recipient must be invoked from the
context, as a definite null complement (Fillmore 1986). Semantically, the
recipient is required to complete the predication. This is one way in which
an argument – in this case the unexpressed recipient – contrasts with
adjuncts. Adjuncts modify the predication rather than complete it, and,
when syntactically absent, the interpretation of the utterance does not
require they be invoked by the listener.
The key evidence that the argument of the applicative -ni (the recipient) is
linked to the undergoer macrorole is the effect of the antipassive suffix, -nVn
(20b) (see Section 6.5.2), which deletes the undergoer.4 With a simple transi-
tive verb in Tepehua, the direct object is syntactically obligatory. Consider
the example sentences in (20). If no object noun phrase appears in the

clause, it is still transitive, requiring a translation with a definite third

singular pronoun (20a).
(20) a. kin-kuku st’a:-y

1poss-uncle sell-ipfv
‘My uncle sells it.’
b. kin-kuku st’a:-nan
1poss-uncle sell-antip
‘My uncle sells.’
c. kin-kuku ki-st’a:-ni-y laqtʃ’iti
1pos-uncle 1obj-sell-dat-ipfv clothes
‘My uncle sells me clothes.’
d. kin-kuku st’a:-ni-nin laqtʃ’iti
1poss-uncle sell-dat-antip clothes
‘My uncle sells clothes’
The translation of (20b) shows that the antipassive suffix -nVn in Tlachichilco
Tepehua results in deletion of the patient undergoer. In (20c), the presence
of the applicative suffix -ni, marks the recipient as undergoer. This is evi-
denced by the fact that it is the recipient that is deleted in (20d), when the
antipassive suffix is present.
In sum, RRG posits only two macroroles even though predicates may have
more than two arguments in their logical structure. Ditransitive construc-
tions display different approaches to the syntactic position of the recipient
argument. In some languages, the theme is the undergoer, following the
default choice on the AUH. In other languages, the recipient is regularly the
marked choice for undergoer, reflecting its prominence as the typically
more animate non-actor argument. Finally, other languages allow variable
linking to the undergoer macrorole.
6.3.3 Other Three-Place Predicates and Lexical Applicatives

While cross-linguistic studies such as Malchukov et al. (2010) limit the
notion of ditransitive to three-place predicates in which the third argument
is some kind of recipient, there are, of course, other kinds of three-place
predicates. Common examples in English include the following.
(21) a. Chris put the jar on the table.

b. Kim sprayed insect repellent under the bed.
The constructions in (21) involve a predicate with an indirect core argument

of location. The predicates put and spray both have the LS [do′ (x, Ø)] CAUSE
[BECOME be-loc′ (y, z)] and ‘in the actual semantic representation of a
sentence be-loc′ would be replaced by the LS of a preposition’ that corres-
ponds to the specific example (e.g. on, under, in, etc.) (Van Valin and LaPolla
1997: 160).
There are at least two ways that the English dative constructions discussed
in 6.3.2 are different from three-place predicates like these. First, these

three-place predicates do not display similar argument alternations that

affect undergoer assignment. Second, they involve predicative prepositions
rather than the non-predicative prepositions found in ditransitives.
There are other classes of three-place verbs in English and other European
languages that do not involve a dative argument but do display alternating
argument structure. One well-known set of such verbs is the spray/load class
(Fillmore 1968: 48; Levin 1993: 50–51).
(22) a. He sprayed the paint on the wall.

b. He sprayed the wall with the paint.
It has often been noted that this alternation reflects a semantic distinction
in that the NP or RP immediately following the verb shows a higher level of
affectedness: as Fillmore notes (1968: 48, fn.49), the sentence in (22b) implies
the entire wall got painted but (22a) does not. Furthermore, in the corres-
ponding passive sentences, the semantic distinction is maintained, showing
that the difference in affectedness is due to undergoer status, not to the
syntactic position of direct object.
(23) a. The paint was sprayed on the wall.

b. The wall was sprayed with the paint.
This alternation is a modulation of argument structure and, like the

dative alternation, reflects different linkings to the undergoer macrorole.
The semantic macrorole of undergoer, not the syntactic position of direct
object, ‘represents the non-instigating, affected participant in a state of
affairs’ (Van Valin 2005: 61–62). Thus, considering the two non-actor argu-
ments in each of the sentences in (22) and (23), the undergoer is the more
affected one (see Van Valin 2005: 113–114).
6.3.4 Causative
Lexical causative constructions built on intransitive verbs are generally
straightforward when it comes to argument structure: the causee is linked
to undergoer in both the intransitive and transitive constructions; in the
transitive construction, the causer links to actor.
(24) a. The door opened.

b. [BECOME open′ (door)]
(25) a. Chris opened the door.

b. [do′ (Chris, Ø)] CAUSE [BECOME open′ (door)]
However, in languages that allow a causative built on a transitive verb base,

the result is a predicate with three core arguments. The causer will be linked
to the actor macrorole, but, as in the case of ditransitives, the question arises
about which of the remaining two arguments will be linked to the
undergoer position.

West Coast Bajau, one of the Sama-Bajaw languages of the Western

Malayo-Polynesian branch of Austronesian, as reported by Miller (2007:
303), has a productive morphological causative prefix pe-.
(26) a. Togor bana tiang pagar e.

upright very post fence dem
‘The fence post stands very straight.’
b. Boi pe-togor Mali tiang pagar e.
comp caus-upright Mali post fence dem
‘Mali erected the fence post.’
Miller (2007: 302–308) includes various examples of causatives with pe- on

statives, inchoatives, manner, and activity verbs. He reports that few transi-
tive verb roots take the pe- causative, but when they do, ‘the causee is the
new undergoer’ (2007: 306). This is also the case in Tepehua, but in a more
indirect fashion. The causee appears in the clause due to the presence of the
same applicative -ni, presented in 6.3.2, which links an indirect object to
the undergoer (the vowel of -ni is lengthened as part of the causative forma-
tion rule).
(27)
transitive causative of transitive
ʔah-ya ‘s/he digs it’ maːʔah-ni:-y ‘s/he makes her/him/it dig it’
ʃʔoq-ya ‘s/he unties it’ ma:ʃʔoq-niː-y ‘s/he makes him/her/it untie it’
In clear contrast to the pattern in the West Coast Bajau and Tepehua
examples is causative formation on a transitive verb in which the undergoer
of the base verb retains the syntactic marking and behaviour of undergoer.
The causee then appears as a peripheral or indirect core argument. An
example of this can be seen in the Turkish sentences in (28) (from Underhill
1976: 346):
(28) a. Yusuf diş-in-i çek-ti

Yusuf tooth-3poss-acc pull-pst
‘Yusuf pulled his tooth.’
b. Yusuf doktor-a diş-in-i c ̣ek-tir-di
Yusuf doctor-dat tooth-3poss-acc pull-caus-pst
‘Yusuf had the doctor pull his tooth.’
In both the basic transitive as well as the causative construction, the direct
object, dis, ‘tooth’ is marked as accusative and, in the causative, the causee
˙
must be marked as dative.
Like other lexical constructions in this section, morphological causatives
affect the level of mapping from predicate argument structure onto the
actor and undergoer macroroles. For causatives formed on a transitive verb
base, two non-actor core arguments are present in the LS, and languages
differ in regard to which of the two links to the undergoer macrorole.

6.3.5 Anticausative
The anticausative construction has been described as the inverse of the
causative: instead of adding a causer and causing event to the LS, the antic-
ausative is a derived intransitive in which the undergoer is the PSA. Haspel-
math and Müller-Bardey (2001) call the anticausative ‘the most radical
agent-removing category’ and include examples such as the following from
Hungarian and Turkish.
(29) a. András-t három tárgy-ból elvág-t-ák

András-acc three subject-elat fail-pst-3pl
‘They failed András in three subjects.’
b. András három tárgy-ból elvág-ód-ott.
András three subject-elat fail-antic-pst(3sg)
‘András failed in three subjects.’
(30) a. Anne-m kapı-yı aç-tı.

mother-1sg door-acc open-pst(3sg)
‘My mother opened the door.’
b. Kapı aç-ıl-dı.
door open-antic-pst(3sg)
‘The door opened.’
In each of these examples, the verb root is causative with actor and under-
goer and the intransitive is derived by a morphological operation.
The RRG literature discusses two kinds of anticausatives. The alternations
in (29) and (30) display an anticausative that is similar to the ‘middle’
construction in traditional grammar. In these constructions, ‘the function
of the morphological markers is to cancel part of the logical structure’ (Van
Valin 2005: 46) – the anticausative removes the causer and causing event
from an otherwise causative verb. Typically, the anticausative alternation
applies to a minor subclass of verbs. This is the case in Tepehua languages,
which have a small group of transitive verb roots that have derived intransi-
tive forms marked by the inchoative prefix, ta-, as in the following examples.
(31)
Base verb Gloss Derived form Gloss
(causative) (anticausative)
laːqaːɬi-y ‘x breaks it down’ ta-laːqaːɬi-y ‘it breaks down’
tʃeʔe-y ‘x shatters it’ ta-tʃeʔe-y ‘it shatters’
ʔeʃ-a ‘x tears it’ ta-ʔeʃ-a ‘it tears’
teʔe-y ‘x cracks it’ ta-teʔe-y ‘it cracks’
The obvious similarity between the passive and this anticausative con-
struction is that both remove the actor as a core argument. However, there is
a key difference between them: the passive has an implicit or understood
actor and causing event but this anticausative construction does not. In RRG
this means that, unlike the passive, in which the actor is implicit but not a

core argument, in these anticausatives the actor and causing event are
missing altogether.
(32) causative achievement/accomplishment ! achievement:

[do′ (x, Ø)] CAUSE [BECOME/INGR pred′ (y)] ! BECOME/INGR pred′ (y)
This kind of anticausative converts a causative achievement or causative

accomplishment with two macroroles into an intransitive achievement or
accomplishment with only the undergoer macrorole.
A second kind of anticausative maintains the causing activity in the LS
but the agent of the activity is unspecified. This kind of anticausative is
found in Romance languages and involves a reflexive construction. The
analysis is presented in Van Valin and LaPolla (1997: 408–414), based on
Centineo (1996), who gives evidence for the presence of the activity
predicate in the LS: the occurrence of manner adverbs – which only
occur with activities – and an implicit agent (see also Bentley 2006:
126–136). Consider these Spanish examples from González Vergara
(2009: 366–374).
(33) a. Pedro ensució la camisa.

Pedro stained.3sg the shirt
‘Pedro stained the shirt.’
b. La camisa se ensució.
the shirt refl stained.3sg
‘The shirt got dirty.’
González Vergara shows that (33b), like the Italian examples in Centineo
(1996), has an implicit causing agent and can occur with manner adverbs.
Thus, unlike the anticausatives accounted for by (32), these maintain the
causing activity in the LS.
(34) Causative achievement/accomplishment ) achievement:

[do′ (x, Ø)] CAUSE [BECOME/INGR dirty′ (y)] )
[do′ (Ø, Ø)] CAUSE [BECOME/INGR dirty′ (y)]
Both of the rules in (32) and (34) change an event with two specified
arguments into an achievement or accomplishment with one undergoer
but they have significantly different logical structures.
6.3.6 Null Complements and the Activity–Active

Accomplishment Alternation
Van Valin (2012: 69) discusses the following examples of the relation
between argument structure and telicity.
(35) a. Sandy wrote (poetry) for an hour/*in an hour. Atelic

b. Sandy wrote the poem in an hour. Telic
(36) a. Chris drank (beer) for an hour/*in an hour. Atelic

b. Chris drank the beer in an hour. Telic

To account for this common alternation between atelic and telic uses of
some verbs, ‘not just verbs but in fact whole verb phrases must be taken
into account to distinguish activities from accomplishments’ (Dowty
1979: 60–62). The RRG analysis of the examples in (35) and (36) requires
distinct logical structures for the atelic and telic readings of the verbs.
The objects in (35a) and (36a) are non-referring NPs but those in (35b) and
(36b) are RPs. As a result, there is a difference in macrorole status, since
activities, including multiple-argument activity verbs (the atelic
examples), never have an undergoer macrorole. A common explanation
for the data in (35) and (36), then, is that the alternation is due to the
change in the referential status of the object, that is, whether it is an RP
or NP.
Van Valin (2012: 69–71) shows that the activity–active accomplishment
alternation in other languages often is not due to the inferred status of the
object, but is marked on the verb. Changing the aspect marking on the verb
can result in a change from activity to active accomplishment (‘He was
eating everything for two hours,’ vs. ‘He ate everything in two hours.’) The
following lexical rules account for the alternations, showing the change in
the LS (see also Van Valin and LaPolla 1997: 180 and, for a revised LS of active
accomplishments which has no consequences for the current discussion,
Van Valin 2018 and Bentley 2019).
(37) a. Activity [motion] ) Active Accomplishment: do′

(x, [pred′ (x)]) ) do′ (x, [pred′ (x)]) & INGR be-LOC′ (y, x)
b. Activity [consumption] ) Active Accomplishment: do′
(x, [pred′ (x, y)]) ) do′ (x, [pred′ (x, y)]) & INGR consumed′ (y)
c. Activity [creation] ) Active Accomplishment: do′
(x, [pred′ (x, y)]) ) do′ (x, [pred′ (x, y)]) & INGR exist (y)
These lexical rules reflect the productivity of an alternation between two LSs
involving the same base verb.
To show that this alternation cannot be attributed simply to the presence of
a definite direct object, Van Valin (2012: 70) presents examples from Georgian
(Holisky 1981), in which the preverb da- imposes a telic reading of the event.
(38) a. K’ac-i (c’eril-s) c’er-s xuti saati.

man-nom (letter-dat) write.prs-3sg five hours
‘The man is writing (letters) for five hours.’
b. K’ac-i c’eril-s da-c’er-s at c’ut-ši.
man-nom letter-dat pv-write.prs-3sg ten minutes-in
‘The man will write the letter in ten minutes’
The kind of alternation accounted for in the rule in (37) is not marked
morphologically in English but is in languages such as Georgian and it
is marked by aspectual markers in other languages such as Russian. The
distinct logical structures of activities and their corresponding active
accomplishments have both semantic and syntactic consequences.

6.4 Syntactic Alternations
This section offers a brief survey of benefactives and constructions which

involve an adjunct. These constructions involve both adjunct prepositions in
the periphery and argument-adjunct prepositions which ‘introduce an argu-
ment into the clause and share it with the logical structure of the core,
rather than taking the logical structure of the core as an argument’ (Van
Valin and LaPolla 1997: 159). These categories of prepositions also character-
ize adpositions and many applicatives cross-linguistically.
6.4.1 Benefactive
Though often treated as a subtype of the dative construction, a benefactive
typically has a different relation to the event referred to by the verb. In the
case of ditransitives, that is, constructions in which the third argument in
some sense ‘receives’ the second one, the verb itself introduces an LS that
includes the three arguments. This is not the case with a benefactive
construction. A benefactive construction involves an argument external
to the event. Unlike the semantic representation of an adjunct, the seman-
tic representation of a benefactive shares an argument with the LS of
the verb.
In RRG, the benefactive involves purpose, and can be represented by the
semantic representation given for ‘purposive for’ in English in Van Valin and
LaPolla (1997: 383) (based on Jolly 1991).
(39) Semantic representation of purposive for:

want′ (x, LS2 ^ DO (x,[LS1 . . .CAUSE. . .LS2])
Like adjunct adpositional phrases, the benefactive has the argument it

licenses as its first argument and embeds the LS of the core as its second
argument. However, the benefactive, unlike an adjunct, shares an argument
with the core. (41) is a simplified analysis of the benefactive example (40).
(40) ki-makaː-ni-ɬ ʔaqa-tawn tʃaqaʔ

1sg.obj-make-dat-pfv clf-one house
‘He made me a house.’
(41) [want′ (x, LS2 ^ DO (x,[make′ (x, house)]) CAUSE [have′ (I, house)]
This benefactive has some similarities to a ditransitive. However, unlike

ditransitives, constructions that include CAUSE and a recipient, benefac-
tives more generally describe a situation of affectedness. Thus, many
languages use the benefactive construction to express malefactives, as well
as with intransitive activities:
(42) ʔik-maqniː-ni-ka-ɬ ki-ʃʔoy

1sbj-kill-dat-pas-pfv 1poss-dog
‘(Someone) killed my dog on me.’

(43) ʔik-miɬpaː-ni-ya-n
1sbj-sing-dat-fut-2obj
‘I will sing for you.’
The benefactive marks the presence of an argument-adjunct that is not part

of the LS associated with the bare verb but, unlike a simple adjunct, it shares
an argument with that LS. Therefore, though it is an example of argument
structure alternation at the syntactic level, it is a lexical process. Indeed, the
Tepehua applicative -ni that marks the presence of the recipient in ditransi-
tives and the experiencer in benefactives is never used to mark the presence
of an adjunct in a clause.
6.4.2 Syntactic Applicatives

Applicative constructions in head-marking languages correspond in large
part to prepositional phrases in dependent-marking languages. This has led
some to analyse applicatives as ‘preposition incorporation’, following Baker
(1988: 229–304). As already shown, the Tepehua applicative dative suffix -ni
results in the indirect object being linked to the undergoer position. This is
also true of the applicative constructions in the following examples from
ˇ
Chichewa (Bantu), as reported in Baker (1988: 247–248).
(44) a. kalulu a-na-gul-ir-a mbidzi nsapato

hare sp-pst-buy-for-asp zebras shoes
‘The hare bought shoes for the zebras.’
b. mbidzi zi-na-gul-ir-idw-a nsapato (ndi kalu!u)
zebras sp-pst-buy-for-pas-asp shoes by hare
‘The zebras were bought shoes by the hare.’
Baker points out that in this construction, the applicative -ir has resulted
in the benefactive argument usurping properties associated with the direct
object: immediate postverbal position, object pro-drop, and (in this
example) subject of the passive (compare 44a and 44b). In RRG terms, this
ˇ
is evidence that in the Chichewa applicative construction, the benefactive
is the undergoer. As we saw in 6.4.1, this is not surprising, as the benefac-
tive is typically an example of an argument-adjunct rather than a
simple adjunct.
However, some applicatives allow adjuncts to appear as syntactic argu-
ments of the verb. As reported in Peterson (2007: 18, 19, 22), Hakha Lai, a
Tibeto-Burman language, has several optional applicative constructions that
allow an adjunct to appear as a verbal argument.
(45) a. ka-law ʔan-ka-thloʔ-pii

1sg.poss-field 3pl.sbj-1sg.obj-weed-com
‘They weeded my field (together) with me.’
b. tiilooŋ khaa tivaa kan-Ø-tan-naak
boat top river 1pl.sbj-3sg.obj-cross-ins
‘We used the boat to cross the river.’

The Hakha Lai examples in (45) have paraphrases in (46):
(46) a. kay-maʔ¼hee ka-law ʔan-thlaw

1sg-pron¼com 1sg.poss-field 3pl.sbj-weed
‘They weeded my field together with me.’
b. tiilooŋ¼ʔin tivaa (khaa) kan-tan

boat¼ins river top 1pl.sbj-cross
‘We used the boat to cross the river.’
Tepehua has three applicative prefixes that allow adjuncts to appear as

arguments of the verb: comitative tʼaː, instrumental puː, and direction ɬi.
(47) a. kin-ta-tʼaː-ʔa-ɬ
1obj-3pl.sbj-com-go-pfv
‘They went with me.’
b. puː-mi-ɬ huːki
ins-come-pfv horse
‘S/he came by horse.’
c. waː yuːtʃa ɬiː-stʼaː-ɬ
foc 3pron dir-sell-pfv
‘S/he sold it for that (price) / sold it for that (reason).’
The instrumental in (47b) has a paraphrase with a Tepehua preposition but

paraphrases of the applicative constructions in (47a) and (47c) require prep-
ositions borrowed from Spanish. (The paraphrase of the comitative requires
kun from Spanish con and the price reading of ɬiː- requires por.)
The two key questions in an RRG account of applicative constructions
built on a transitive verb base involve the lexical level (48a) and the syntactic
level (48b).
(48) a. Does the argument of the base verb or the argument of the applicative link
to the undergoer macrorole?
b. Is there a restriction regarding which of the non-actor arguments may
occur as PSA in a passive construction?
The question in (48b) is touched on in Section 6.5. Data regarding (48a) is

provided by Peterson (2007) regarding applicatives in Bakusu (Bantu) and
Hakha Lai (Tibeto-Burman). In Bakusu, the instrumental applicative does
not manifest standard object properties and a first hypothesis would be
that the instrument does not link to undergoer position though it is a
syntactic argument of the derived verb. In Hakha Lai, in six of the seven
applicative constructions the applied object apparently links to undergoer
position, taking on standard object properties. The one exception, again, is
the instrumental, a feature that Peterson suggests is due to the fact that
instrumentals are typically inanimate, and, therefore, less salient in
reported events.
In Tlachichilco Tepehua, only the argument of the dative applicative,
-ni, is regularly linked to the undergoer macrorole. The other three

applicatives, ɬiː-, puː-, tʼaː-, most often simply allow adjuncts to appear as
syntactic arguments of the verb – an important feature since only direct
arguments of the verb can be questioned or relativized in Tepehua.
However, the linking of the PSA in a passive construction is not limited
to the undergoer: any direct syntactic argument of the verb can be
the PSA.
6.5 Voice
The RRG account of voice alternations – passive and antipassive – distin-

guishes two areas of argument linking. RRG’s universal formulation of voice
oppositions refers to both the syntactic and lexical dimensions of voice
modulation.
(49) General characterization of basic voice constructions in RRG5

a. PSA modulation voice: permits an argument other than the default
argument . . . to function as the privileged syntactic argument.
b. Argument modulation voice: gives non-canonical realization to a
macrorole argument.
(Van Valin 2005: 116)
In RRG, the general cross-linguistic characterization of passive and anti-

passive voice involves both levels of mapping: the lexical assignment of
macrorole status and the syntactic determination of PSA. Van Valin and
LaPolla (1997) provide the English passive and the Dyirbal (a Pama-Nyugan
language spoken in north-eastern Australia) antipassive as prototypical
constructions.
(50) a. English passive construction

b. Dyirbal -ŋay antipassive construction
Though these are given as prototypical examples, it should be noted that

there are languages with passive or antipassive constructions in which
modulation occurs on only one of the two levels mentioned in (49).
As a result of argument modulation, in a passive construction, the actor
macrorole is removed from the core – it is either deleted or moved to the
periphery. This typically results in some other argument occurring as the
PSA. In the case of the antipassive, the undergoer is similarly removed,
typically making an otherwise transitive verb into an intransitive.
6.5.1 Passive
According to Keenan and Dryer (2007: 328–329), the following passives are
‘basic passives’ ((51) is their example, (52) is Tlachichilco Tepehua).
(51) John was slapped.

(52) Pa:laqsti-saː-ka-ɬ ni Juan

cheek-hit-pass-pfv art John
‘John was slapped.’
This is considered to be a ‘basic’ passive because passives like this ‘are the
most widespread across the world’s languages’. They have the following
characteristics: (i) there is no agent phrase (in Tepehua, as in many other
languages, the passive does not allow the actor to appear in the clause at
all); (ii) the verb is a transitive verb that is passivized; (iii) in its non-
passivized form, the verb expresses an action with an agent subject and
patient object.
By these criteria, then, English passives with an explicit actor as well as
impersonal passives (passives of intransitives) are not basic passives. The
contrast between the two Tepehua forms in (53) illustrates the use of the
passive with intransitive verbs, as do the examples from German and
Turkish in (54a) and (54b), respectively.
(53) a. Tapaːtsaː-kan maɬkuyuː abril y mayu

work-pass(ipfv) month April and May
‘It is worked (people work) the months of April and May.’
b. ʔantʃa ʔalin-kan
there exist-pass(ipfv)
‘It is existed there./Something is there.’
(54) a. Es wird hier getanzt

it is here danced.
‘Dancing takes place here.’
b. Eğlen-il-di.
have.fun-pass-pst
‘Fun was had.’
The existence of impersonal passives shows the importance of distinguish-

ing the two aspects of passive in (49). The forms in (53) and (54) involve the
argument modulation of (49b), in that the actor is absent from the clause.
However, the PSA modulation is irrelevant in these constructions, as there is
no PSA.
Even very closely related languages can have passive constructions that
differ in either the PSA modulation or the argument modulation. In Tlachi-
chilco Tepehua, first- and second-person undergoers (but not third plural)
are marked as PSA in the passive construction (55a). In Pisaflores Tepehua,
while second-person undergoers must be marked as PSA, first-person under-
goers more commonly retain the undergoer marking (55b).
(55) a. k-laqtsʼin-kan-a:-w
1sbj-see-pass-ipfv-1pl
‘We are seen.’
b. kin-ta-laqts’in-kan-a:-n
1obj-3plsbj-see-pass-ipfv-2obj
‘We are seen.’

In the Tlachichilco Tepehua example, (55a), the actor is removed (argument

modulation) and the undergoer is the PSA, marked by the ‘subject’ form
(PSA modulation). In the Pisaflores Tepehua example (55b), the actor is
removed (argument modulation) but the undergoer is still marked as
‘object’, not as PSA (there is no PSA modulation).
Besides the variation regarding the presence or absence of PSA modula-
tion, there is notable cross-linguistic variation regarding what arguments
can be linked to the PSA.
Yaqui (Uto-Aztecan) is an example that only allows a macrorole to appear
as PSA, so in the following passive forms, only the undergoer can function as
PSA (from Guerrero and Van Valin 2004: 299–300).
(56) a. U chu’u-W ki’i-wa-k.

the dog-nom bite-pass-pst.pfv
‘The dog was bitten.’
b. Jamut-ta-u nooka-wa-k.
woman-acc-dir talk-pass-pst.pfv
‘Someone talked to the woman’ / *‘The woman was talked to.’
In the passive construction in (56a) the undergoer assumes PSA status; in the
passive in (56b), however, the woman is marked by the directional postpos-
ition, -u, and is not linked to the undergoer position. As a result, the PSA
modulation of the passive does not apply, only the argument modulation,
removing the actor.
However, some languages allow a wide variety of arguments to link to
the PSA position. Van Valin (2005: 121) gives examples from Kinyarwanda
(Bantu), in which applicatives can mark several semantic roles. Tepehua is
another example. The Totonac-Tepehua languages have pragmatically
determined word order and no case marking on the NPs. It is not possible
to determine, for a transitive clause in isolation with two third-person
singular participants, which is the PSA. The passive and antipassive play
a major role in tracking referents. In (57), the instrumental and comitative
appear as syntactic arguments of the verb due to the applicatives though
they do not have macrorole status. However, in the passive they can map
onto the PSA.
(57) a. waː yuːtʃa puː-tʃʼan-nan-kan

foc that ins-sow-antip-pass(ipfv)
‘That’s what it’s planted with (instrumental).’
b. ʔik-tʼaː-tʃiwin-ka-ɬ
1sbj-com-speak-pass-pfv
‘I was spoken with (comitative).’
In Tepehua, the discourse-pragmatics directly influence the selection of

the PSA, and the freedom for different arguments to function as PSA serves a
key pragmatic function: maintaining topics and tracking referents (see
Watters 2017).

6.5.2 Antipassive
The term ‘antipassive’ refers to a change in syntactic alignment of argu-
ments that has traditionally been used to describe a detransitivizing form in
ergative languages, such as in the following examples from Dyirbal (from
Dixon 1994: 161, 170).
(58) a. yabu banaga-nyu

mother.abs return-nonfut
‘Mother returned.’
b. ŋuma yabu-ŋgu bura-n
father.abs mother-erg see-nonfut
‘Mother saw father.’
c. yabui [bural-ŋa-ŋu ŋuma-gu] banaga-nyu
mother.abs see-antip-rel.abs father-dat return-nonfut
‘Mother, who saw father, was returning.’
The examples in (58a) and (58b) exemplify the standard ergative-absolutive

marking for actor in an intransitive clause and in a transitive clause. The
absolutive is the PSA in Dyirbal and only the PSA can be relativized. As a
result, unlike nominative-accusative languages, for the actor to be relativ-
ized, it must be absolutive. The sentence in (58c) shows how this can be
done. The antipassive suffix on the verb in the relative clause results in an
intransitive construction, the undergoer is no longer a core argument and
the actor, as with any intransitive, is in the absolutive, allowing the forma-
tion of the relative clause.
Much of the discussion in the RRG literature regarding antipassives
involves syntactically ergative languages such as Dyirbal. In those cases,
the antipassive construction allows the actor to function as PSA. Some
linguists use the term antipassive only for ergative languages. However,
I am following the perspective articulated by Polinsky (2013):
Some researchers insist on the link between the antipassive and

ergativity . . ., while others propose that the antipassive is not limited to
ergative languages . . . The transitive/antipassive alternation is simply
more visible in an ergative language, where it typically involves a
change in subject case marking from ergative to absolutive.
Considering the two aspects of voice in RRG (49), the same can be said for
antipassives as for passives: in some languages the relevant construction
only affects the argument modulation and does not directly affect the PSA
modulation. In fact, many ergative languages are only ergative in their
morphology and do not have the syntactic ergativity found in a language
like Dyirbal.
In a non-ergative language like Tepehua, the antipassive only involves
the level of argument modulation (49b). It occurs on a transitive verb base,
marking the absence of the undergoer, whether it is the undergoer of the
verb root or the marked undergoer, that is, the argument of the dative

suffix -ni. But it never marks the absence of the applicative argument
associated with one of the applicative prefixes (comitative, instrumental,
directional).
(59) a. puː-stʼaː-na-ɬ ki-muːral

ins-sell-antip-pfv 1poss-bag
‘S/he sold using my bag.’
b. José tʼaː-stʼaː-na-ɬ Kwan
José com-sell-antip-pfv Kwan
‘José sold with Kwan.’
These examples involve the antipassive construction, which marks the

absence of the undergoer. However, the arguments of the instrumental and
comitative applicatives are present. This supports the analysis in 6.4.2 that
Tepehua applicatives other than the dative -ni, do not link to undergoer.
6.6 Conclusion
Alternations in argument structure at the lexical and syntactic levels have

important functions. Some lexical processes such as causatives and anticausa-
tives are often more limited in productivity and serve to enrich the lexicon.
Others, such as noun incorporation and ditransitive alternations, and some
applicatives, clearly have semantic effects regarding which argument is the
most affected or patient-like (i.e. the undergoer). The activity–active accom-
plishment alternation involves a change in transitivity and is tied to a
difference in the telic or non-telic nature of the event.
Syntactic processes discussed in this chapter add syntactic arguments to the
verb or affect the linking to the PSA. Voice alternations, such as passive and
antipassive, involve mappings at both the lexical and syntactic levels. Apart
from simply removing an argument from the core because that argument is
not salient at the moment, these constructions often play an important role
in tracking referents or maintaining the discourse topic.
RRG provides a framework that has the heuristic value of requiring one to
determine the logical structure associated with a verb and the linking of
arguments in the logical structure to the macroroles and from the macro-
roles to PSA and syntactic positions in the clause. The alternations discussed
in this chapter are captured by capitalizing on the distinction between these
levels of linguistic analysis.
References
Allen, Barbara J., Donna B. Gardiner and Donald G. Frantz. 1984. Noun
incorporation in Southern Tiwa. Indonesian Journal of Applied Linguistics
50: 292–311.

Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing.

Chicago: University of Chicago Press.
Bentley, Delia. 2006. Split Intransitivity in Italian. Berlin: Mouton de Gruyter.
Bentley, Delia. 2019. The Logical Structure of Verbs of Quantized and Non-Quantized
Change. Paper presented at the International RRG Conference, 19–21
August 2019, University at Buffalo (SUNY).
Centineo, Giulia. 1996. The distribution of si in Italian intransitive/incho-
ative pairs. In Mandy Simmons and Theresa Galloway (eds.), Proceedings of
SALT V, 54–71. Ithaca, NY: Cornell University Press.
Dixon, R. M. W. 1994. Ergativity. Cambridge: Cambridge University Press.
Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: D.
Reidel.
Dryer, Matthew S. 1986. Primary objects, secondary objects, and antidative.
Language 62: 808–845.
Fillmore, Charles J. 1968. The case for case. In Emmon Bach and Robert T.
and Winston.
Fillmore, Charles J. 1986. Pragmatically controlled zero anaphora. In Vassi-
liki Nikiforidou, Mary VanClay, Mary Niepokuj and Deborah Feder (eds.),
Twelfth Annual Meeting of the Berkeley Linguistics Society, 95–107. Berkeley, CA:
Berkeley Linguistics Society.
Foley, William A. and Robert D. Van Valin. 1984. Functional Syntax and Univer-
sal Grammar. Cambridge: Cambridge University Press.
González Vergara, Carlos. 2009. One rule to rule them all: Logical structures
for Spanish non-reflexive se sentences. In Lilián Guerrero, Sergio Ibáñez
Cerda and Valeria A. Belloro (eds.), Studies in Role and Reference Grammar,
361–380. Mexico: Universidad Nacional Autónoma de México.
Guerrero, Lilián and Robert D. Van Valin. 2004. Yaqui and the analysis of
primary object languages. International Journal of American Linguistics 70(3):
290–319.
Haspelmath, Martin and Thomas Müller-Bardey. 2001. Valency change. In
G. Booij, C. Lehmann and J. Mugdan (eds.), Morphology. An International
Handbook on Inflection and Word Formation, 207. Berlin: Walter de Gruyter.
www.eva.mpg.de/fileadmin/content_files/staff/haspelmt/pdf/2005val.pdf.
Holisky, Dee A. 1981. Aspect theory and Georgian aspect. In B. Comrie and M.
Polinsky (eds.), Causatives and Transitivity, 87–120. Amsterdam: John
Benjamins.
Jackendoff, Ray S. 2002. Foundations of Language: Brain, Meaning, Grammar,
Evolution. Oxford: Oxford University Press.
Jolly, Julia. 1991. Prepositional Analysis within the Framework of Role and Reference
Grammar. New York: Peter Lang.
Keenan, Edward L. and Matthew S. Dryer. 2007. Passive in the world’s
languages. In Timothy Shopen (ed.), Language Typology and Syntactic Descrip-
tion, Vol. 1: Clause Structure, 325–361. Cambridge: Cambridge University
Press.

Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary Investi-
gation. Chicago: University of Chicago Press.
Malchukov, Andre, Martin Haspelmath and Bernard Comrie. 2010. Studies
in ditransitive constructions: A typological overview. In Andre Malchukov,
Martin Haspelmath and Bernard Comrie (eds.), Studies in Ditransitive Con-
structions. A Comparative Handbook, 1–64. Berlin: De Gruyter, Mouton.
McGregor, William. 1997. Grammatical structures in noun incorporation. In
Anne-Marie Simon-Vandenbergen, Kristin Davidse and Dirk Noël (eds.),
Reconnecting Language: Morphology and Syntax in Functional Perspectives. Cur-
rent Issues in Linguistic Theory, 141–180. Amsterdam: John Benjamins.
Miller, Mark. 2007. A Grammar of West Coast Bajau. PhD dissertation, (Arling-
ton), University of Texas.
Mithun, Marianne. 1984. The evolution of noun incorporation. Language 60:
847–894.
Perlmutter, David M. and Paul M. Postal. 1977. Toward a universal charac-
terization of passive. In K. Whistler, R. D. Van Valin et al. (eds.), Proceedings
of the Annual Meeting of the Berkeley Linguistics Society, 3, 394–417. Berkeley,
CA: Berkeley Linguistics Society.
Peterson, David A. 2007. Applicative Constructions. Oxford: Oxford University
Press.
Polinsky, Maria. 2013. Antipassive constructions. In Matthew S. Dryer and
Martin Haspelmath (eds.), The World Atlas of Language Structures Online.
Leipzig: Max Planck Institute for Evolutionary Anthropology. http://wals
.info/chapter/108.
Rosen, Sarah Thomas. 1989. Two types of noun incorporation: A lexical
analysis. Language 65: 294–317.
Sadock, Jerrold M. 1991. Autolexical Syntax: A Theory of Parallel Grammatical
Representations. Chicago: University of Chicago Press.
Sapir, Edward. 1911. The problem of noun incorporation in American lan-
guages. Language 13(2): 250–282.
Underhill, Robert. 1976. Turkish Grammar. Cambridge: MIT Press.
Van Valin, Robert D. Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
Van Valin, Robert D. Jr. 2007. The Role and Reference Grammar analysis of
three-place predicates. Suvremena Lingvistika 33.1(63): 31–64. https://rrg.caset
.buffalo.edu/rrg/vanvalin_papers/RRG-Analysis_Three-Place_Pred.pdf.
Van Valin, Robert D. Jr. 2012. Lexical representation, co-composition, and
Generative Lexicon Theory: Text, Speech and Language Technology, Vol. 46,
67–107. Dordrecht: Springer.
Van Valin, Robert D. Jr. 2018. Some issues regarding (active) accomplish-
and Expanding Role and Reference Grammar, 71–93. Freiburg: Freiburg Insti-
tute for Advanced Studies, Albert-Ludwigs Universität Freiburg.

Van Valin, Robert D. Jr and Randy LaPolla. 1997. Syntax: Structure, Meaning,
Vendler, Zeno. 1967[1957]. Linguistics in Philosophy. Ithaca, NY, Cornell Univer-
sity Press. (Previously published in Vendler, Zeno. 1957. Verbs and times.
Philosophical Review 56: 143–160.)
Watters, James K. 2017. Tlachichilco Tepehua: Semantics and function of
verb valency change. In Álvaro González and Ía Navarro (eds.), Verb Valency
Changes: Theoretical and Typological Perspectives, Typological Studies in Lan-
guage 120, 165–192. Amsterdam: John Benjamins.
Watters, James K. 2019. La preposición en tepehua y construcciones seme-
jantes. In Lilián Guerrero (ed.), Adposiciones y elementos de su tipo en lenguas
de América. Estudios sobre Lenguas Americanas 9, 315–344. Mexico: Uni-
versidad Nacional Autónoma de México, Instituto de Investigaciones
Filológicas.
Williams, Alexander. 2015. Arguments in Syntax and Semantics. Cambridge:
Zylstra, Carol F. 1991. A syntactic sketch of Alacatlazala Mixtec. In Bradley, C.
Henry and Barbara E. Hollenbach (eds.), Studies in the Syntax of Mixtecan
Languages, Vol. 3, 1–178. Dallas: Summer Institute of Linguistics and the
University of Texas at Arlington.
Notes
1 When citing research done by others, I have tried to keep the same
morpheme glosses that they have used.
2 For discussion of body-part incorporation with an intransitive verb,
resulting in a change of undergoer assignment (‘possessor raising’), see
Van Valin (2005: 145–146).
3 The gloss B refers to one of three gender classes, in this case, one that
refers to either an animate plural or an inanimate singular (Allen et al.
1984: 293, fn. 5)
4 The passive or ‘unspecified subject’ construction in Tepehua does not
provide evidence of undergoer status. In a Tepehua passive construction,
a non-undergoer can map onto the PSA (see 6.4.2).
5 These parallel the ‘two universals of passivization’ presented by Perlmut-
ter and Postal: (i) ‘A direct object of an active clause is the (superficial)
subject of the “corresponding” passive,’ and (ii ‘The subject of an active
clause is neither the (superficial) subject nor the (superficial) direct object
of the “corresponding” passive’ (1977: 399).

7
Case Assignment
Wataru Nakamura
Abbreviations

ACT actor LS Logical Structure

ADE adessive MP modifier phrase
AV actor voice MR macroroe
BV benefactive voice NUC nuclear
C catalyst particle OT Optimality Theory
CC Completeness Constraint PRT partitive
DCA domain of case PSA privileged syntactic
assignment argument
DMAP default macrorole PURP purpose
assignment principles
ELA elative PV patient voice
GER gerund QS qualia structure
ICR instrumental case rule REC.PFV recent perfective
ILL illative RP reference phrase
INE inessive UND undergoer
IO indirect object
7.1 Introduction
The aim of this chapter is twofold: to explicate the theory of case assignment
in Role and Reference Grammar (RRG; Van Valin 1991, 1993, 2005, 2009; Van
Valin and LaPolla 1997) and to outline its Optimality-Theoretic (OT) imple-
mentation (Nakamura 1999a, 1999b) and its extension to instrumental case
assignment and case syncretism (Nakamura 2011, 2021).
Case in RRG constitutes part of the linking system in which verbal argu-
ments are realized by case/cross-referencing markers, agreement markers,

Case Assignment 319
and/or word-order positions, but it plays no direct syntactic role in RRG,

unlike in Government and Binding (GB)/Minimalism. Another distinguish-
ing feature of the RRG theory of case assignment is that it ties the core cases
(i.e. those that mark A, O and S arguments) to macrorole status (instead of
identifying semantic roles of arguments they mark or assigning them to
either particular grammatical relations or structural positions) and treats
dative case on a par with those core cases by defining it as the default case
for non-macrorole core arguments (Van Valin 1991; cf. Silverstein 1980/
1993). What is notable about the macrorole-dependent theory of case assign-
ment in RRG is that it makes no reference to any phrase-structural positions
and therefore is well equipped to handle case-marking systems in both
configurational and non-configurational languages.
There are five major features of the RRG theory of case assignment as
outlined in Van Valin and LaPolla (1997) and Van Valin (1993, 2005). First,
RRG assigns the core cases (i.e. nominative, accusative, absolutive and erga-
tive) in terms of the ranking of macroroles instead of tying them to gram-
matical relations or structural positions (cf. Yip et al. 1987; Marantz 1991;
Baker 2015). Second, unlike GB/Minimalism, RRG does not assume that case
assignment takes place in all languages whether or not they have any overt
case markers; it does not distinguish between syntactic cases and morpho-
logical cases and assigns case markers directly to NPs (see Section 7.5 for an
alternative proposal). Third, RRG adopts the macrorole-based definition of
dative case and treats it on a par with the core cases. Fourth, RRG treats
verbal cross-reference and nominal case systems in a unified way due to
their functional equivalence. Finally, RRG parameterizes the domain of case
assignment (DCA) into the core and clause and allows the set of case
assignment rules to apply in each core independently or to all of the cores
in a clause jointly. Taken together, these features distinguish RRG from the
other syntactic frameworks with respect to case assignment.
This chapter is organized as follows. Section 7.2 provides a brief summary
of the RRG linking system, highlighting those aspects of it that are relevant
to case assignment. Section 7.3 presents the RRG account of accusative,
ergative, and active(-stative) (or split-S) case systems and summarizes the
RRG account of oblique case assignment with a particular focus on dative
and instrumental case. Section 7.4 outlines the OT implementation of the
RRG theory of case assignment (Nakamura 1999a, 1999b) and its extension to
instrumental case (Nakamura 2021). Section 7.5 introduces the distinction
between syntactic and morphological cases into RRG to develop the OT-RRG
account of case syncretism. Section 7.6 concludes the chapter.
7.2 Linking Theory in RRG
7.2.1 Syntactic Structure

RRG is a monostratal theory that posits only a single syntactic representa-
tion for a sentence that consists of two projections, the constituent structure

320 WATARU NAKAMURA
CLAUSE
CORE PERIPHERY
RP NUC RP MP
PRED
V ADV
John saw Tom yesterday

Figure 7.1 The constituent structure of the simple clause in English (RP ¼ reference phrase;
MP ¼ modifier phrase)
projection and the operator projection. The former projection represents the
clause structure with three nested layers, the nucleus (the predicate), the
core (the predicate and its arguments), and the clause (the core and any
peripheral elements that modify the core) (as illustrated in Figure 7.1), while
the latter consists of auxiliary elements (e.g. aspect, tense, root/epistemic
modal, status, evidential, speech act) that are hierarchically and topologic-
ally ordered according to the layer they modify.
7.2.2 Semantic Representation and Syntactic Function

The above syntactic representation is coupled with the semantic representa-
tion of the clause based on the following decompositional representations of
predicates (termed logical structures (LS)), adapted from Vendler (1967) and
Dowty (1979) (Van Valin 2005: 42–49).
(1) Decompositional representations for Aktionsart classes

a. State predicate′ (x) or (x, y)
b. Activity do′ (x, [predicate′ (x) or (x, y)]), or
c. Achievement INGR predicate′ (x) or (x, y), or
INGR do′ (x, [predicate′ (x) or (x, y)])
d. Semelfactive SEML predicate′ (x) or (x, y), or
SEML do′ (x, [predicate′ (x) or (x, y)])
e. Accomplishment BECOME predicate′ (x) or (x, y), or
BECOME do′ (x, [predicate′ (x) or (x, y)])
f. Active accomplishment
do′ (x, [predicate1′ (x) or (x, y)]) & INGR predicate2′ (z, x) or (y)
g. Causative α CAUSE β, where α and β are logical structures of any type
(2a)–(2g) illustrate the above decompositional representations.
(2) a. State
The cup is shattered. shattered′ (cup)
Carl is in the library. be-in′ (library, Carl)

Case Assignment 321
b. Activity
The children cried. do′ (children, [cry′ (children)])
Kim ate fish. do′ (Kim, [eat′ (Kim, fish)])
c. Achievement
The windows shattered. INGR shattered′ (windows)
The balloon popped. INGR popped′ (balloon)
d. Semelfactive
Dana glimpsed the dog. SEML see′ (Dana, dog)
Mary coughed. SEML do′ (Mary, [cough′ (Mary)])
e. Accomplishment
The snow melted. BECOME melted′ (snow)
Chris learned French. BECOME know′ (Chris, French)
f. Active accomplishment
John ran to the park. do′ (John, [run′ (John)]) & INGR be-at′ (park, John)
Kim ate the fish. do′ (Kim, [eat′ (Kim, fish)]) & INGR consumed′ (fish)
g. Causative
The dog scared the boy. [do′ (dog, ø)] CAUSE [feel′ (boy, [afraid′])]
Max melted the ice. [do′ (Max, ø)] CAUSE [BECOME melted′ (ice)]
Felix bounced the ball. [do′ (Felix, ø)] CAUSE [do′ (ball, [bounce′ (ball)])]
A key component of the RRG linking system is the two-tiered system of

semantic roles. The first tier is thematic relations such as effector, experi-
encer, theme and patient. They are defined in terms of argument positions
in the decompositional representations. The second tier comprises two
semantic macroroles (MR), actor (ACT) and undergoer (UND). These are
generalized semantic roles that subsume a number of LS arguments for
morphosyntactic purposes and correspond to the two primary arguments
of a transitive verb.1
The number and nature of macroroles that a verb takes is determined by
the default macrorole assignment principles (DMAP) in (3).

a. Number: the number of macroroles which a verb takes is less than or equal
to the number of arguments in its LS:
b. Nature: for verbs which take one macrorole:
(3a) determines the number of macroroles a verb takes, while (3b) deter-
mines which macrorole (actor or undergoer) it is when the verb receives only
one macrorole. When the number of macroroles does not follow from the
DMAP, it has to be specified in the lexical entry of the verb by a feature [MRα]
(where α represents the number of macroroles).
The relationship between LS argument slots (or thematic relations serving
as mnemonics for them) and macroroles is captured by the Actor-Undergoer
Hierarchy (AUH) in Figure 7.2.

322 WATARU NAKAMURA
Actor Undergoer
Arg. of Arg. of 1st Arg. of 2nd Arg. of Arg. of state

DO do′ (x,...) pred′ (x, y) pred′ (x, y) pred′ (x)
Agent Effector Locative Theme Patient
Experiencer
[ = increasing markedness of realization of LS argument as macrorole]
Figure 7.2 Actor-Undergoer Hierarchy (adapted from Van Valin and LaPolla 1997: 146)
The AUH states that given the LS of a multi-argument verb, the leftmost
argument will be the actor and the rightmost argument will be the
undergoer.
(4a)–(4f) illustrate the default macrorole assignment.
(4) a. John [Effector, ACT] killed Bill [Patient, UND].

b. Bill [Patient, UND] was killed by John [Effector, ACT].
c. John [Effector, ACT] gave a book [Theme, UND] to Bill [Recipient, Non-MR].
d. John [Experiencer, ACT] knew the student [Theme, UND].
e. John [Effector, ACT] ran to the park.
f. John [Patient, UND] disappeared suddenly.
It is important to note that marked assignments to undergoer are possible,

as illustrated by the locative alternation in (5a,b), in which the locative
argument (the truck) may be chosen as undergoer in violation of the AUH.
(5) a. John loaded hay on the truck.

b. John loaded the truck with hay.
The observation that the locative argument in (5b) (the truck) is construed as
being fully loaded motivates its choice as undergoer.2
Finally, the privileged syntactic argument (PSA) selection is based on the
PSA Selection Hierarchy in (6), the PSA selection principles in (7), and
restrictions in (8) (Van Valin 2005: 100).
(6) PSA Selection Hierarchy

arg. of DO > 1st arg. of do′ > 1st arg. of pred′ (x, y) > 2nd arg. of pred′ (x, y) >
arg. of pred′ (x)
(7) Privileged syntactic argument selection principles

a. Accusative constructions: highest-ranking direct core argument in terms
of the PSA Selection Hierarchy (default)
b. Ergative constructions: lowest-ranking direct core argument in terms of
the PSA Selection Hierarchy (default)
(8) Restrictions on PSA in terms of macrorole status

a. Languages in which only macrorole arguments can be PSA (e.g. German,
Sama, Dyirbal, Jakaltek)
b. Languages in which non-macrorole direct core arguments can be PSA (e.g.
Icelandic, Georgian, Japanese, Kinyarwanda)

Case Assignment 323
Since an actor argument always outranks an undergoer argument in the

PSA Selection Hierarchy, (7a) groups A and S arguments against
O arguments, while (7b) groups O and S arguments against A arguments.
7.3 The Theory of Case Assignment
7.3.1 Regular Cases

This subsection is a summary of the RRG theory of case assignment outlined
in Van Valin and LaPolla (1997) and Van Valin (2005). (9) and (10) are the sets
of case assignment rules for accusative and ergative case systems, respect-
ively. They refer crucially to the PSA Selection Hierarchy in (6), and not to
grammatical relations or phrase structural positions.
(9) Case assignment rules (Accusative)

a. Assign nominative case to the highest-ranking macrorole argument.
b. Assign accusative case to the other macrorole argument.
c. Assign dative case to non-macrorole core arguments (default).
(10) Case assignment rules (Ergative)

a. Assign absolutive case to the lowest-ranking macrorole argument.
b. Assign ergative case to the other macrorole argument.
‘The highest-ranking macrorole argument’ in (9a) conflates A and

S arguments, while ‘the lowest-ranking macrorole argument’ in (10a) con-
flates O and S arguments. What is noteworthy about (9) and (10) is that they
define all the core cases (i.e. nominative, absolutive, accusative and ergative)
and dative case with reference to (non-)macrorole status and, in so doing,
treat dative as one of the regular cases along with the core cases.3
Defining dative as the default case for non-macrorole core arguments
accounts for why source arguments of ditransitive verbs may receive dative
case as well as ablative case in many languages and leaves room for its being
overridden by other oblique cases that denote a more specific semantic
content in terms of LS configuration.4 For example, (11a) is the ablative
preposition assignment rule for English, which applies to the uses of from
in (11b).
(11) a. MR: Non-MR

LS: the first argument in the following LS configuration
‘. . . BECOME/INGR NOT have′/be-LOC′ (x, y)’
b. 1. John came from Chicago.
2. John received the watch from Mary.
(11a) is more specific than the macrorole-based definition of dative case in

(9c) and (10c), in that it refers to a particular LS argument in addition to its
non-macrorole status. This accounts for why the ablative preposition is used
to mark source arguments as illustrated in (11b).5

324 WATARU NAKAMURA
The following shows how (9) and (10), respectively, work in Icelandic and
Warlpiri and how the macrorole-based account of case assignment extends to
active(-stative) case systems. First, let us consider the Icelandic data given in (12).
(12) Icelandic (Andrews 1990: 188; Van Valin 1991: 172)

a. Stelpurnar hláu.
the.girls.nom laughed
‘The girls laughed.’
b. Lögreglan tók Siggu fasta.
the.police.nom took Sigga.acc fast.acc
‘The police arrested Sigga.’
c. Stelpan sýndi stráknum myndavélina.
the.girl.nom showed the.boy.dat the.camera.acc
‘The girl showed the boy the camera.’
d. Ég tel lögregluna hafa tekið Siggu
I.nom believe the.police.acc have taken Sigga.acc
fasta.
fast.acc
‘I believe the police to have arrested Sigga.’
(12a)–(12c) illustrate canonical intransitive, transitive and ditransitive con-

structions, while (12d) illustrates matrix-coding (‘raising’) constructions in
which the PSA of the dependent core shows up in the matrix core. The case
assignment in (12a)–(12d) proceeds as follows. The actor arguments in (12a)–
(12c) correspond to ‘the highest-ranking macrorole argument’ in (9a) and
receive nominative case. The undergoer arguments in (12b,c) correspond to
‘the other macrorole argument’ in (9b) and take accusative case. Likewise, the
experiencer argument in the matrix core of (12d) receives nominative case
from (9a), while both the matrix-coded argument lögreglan ‘the police’ and the
patient argument Sigga in the dependent core correspond to ‘the other
macrorole argument’ in (9b) and receive accusative case.
Next, let us consider how to account for the DAT-NOM and NOM-DAT-DAT
case frames in (13a)–(13c), in which the macrorole assignment does not
follow straightforwardly from the DMAP in (3).
(13) Icelandic (Andrews 1990: 210; Van Valin 1991: 174, 175)
a. Stráknum líkar slíkir bílar.
the.boy.dat likes such cars.nom
‘The boy likes such cars.’
b. Henni hefur alltaf þótt Ólafur leiðinlegur.
her.dat has always thought Olaf.nom boring.nom
‘She has always considered Olaf boring.’
c. Ég skilaði henni peningunum.
I.nom returned her.dat the.money.dat
‘I returned her the money.’
The DMAP predicts that all of the multiple-argument verbs in (13a)–(13c) take
a pair of actor and undergoer. However, this prediction is not borne out;
contrary to (3a1), all of (13a)–(13c) contain only one macrorole argument.

Case Assignment 325
In contrast to the earlier analyses of quirky case in Icelandic (e.g. Andrews

1982, 1990; Yip et al. 1987), RRG derives the irregular (quirky) case marking
illustrated in (13a)–(13c) from the irregularity in their macrorole transitiv-
ity. Specifically, Van Valin (1991) posits that þykja ‘think’, líka ‘like’, and skila
‘return’ in (13) are lexically prespecified for having only one macrorole
([MR1]) despite having two arguments in their LS.
Let us see how this lexical pre-specification allows us to treat dative as one
of the regular cases (Van Valin 1991; cf. Narasimhan 1998; Nakamura 1999b,
2008). First, suppose that the two-place state verb líka ‘like’ receives only one
macrorole because of its lexical specification. (3b) dictates that the only
macrorole is an undergoer, since the verb has no activity predicate in its
LS. The AUH requires the theme argument slíkir bílar ‘such cars’ to become
an undergoer. This leads the experiencer argument to receive a non-
macrorole status.
(14) Macrorole assignment of the verb líka ‘like’ ([MR1])

MR Non-MR Undergoer
LS: like′ (boy, such cars)
The case assignment rules in (9) apply to the combination of the non-
macrorole and undergoer arguments and yield the DAT-NOM case frame
in (13a). The same account holds for the DAT-NOM case frame in (13b).
Likewise, suppose that skila ‘return’ in (13c) is lexically specified for having
only one macrorole ([MR1]). (3b) requires that the only macrorole that the
verb takes is an actor, since it has an activity predicate do′ in its LS in (15).
(15) LS of the verb skila ‘return’

[do′ (x, ø)] CAUSE [INGR have′ (y, z) ^ NOT have′ (x, z) ] [MR1]
The effector argument is an actor, but the remaining non-PSA arguments

have no choice but to receive a non-macrorole status, as shown in (16):
(16) Macrorole assignment of the verb skila ‘return’ ([MR1])

MR Actor Non-MR Non-MR
LS: [do′ (I, ø)] CAUSE [INGR have′ (she, money) ^ NOT have′ (I, money)]
This macrorole assignment accounts for why the non-PSA arguments in (13c)
receive dative case. Lexical specification of the number of macroroles that
þykja ‘think’, líka ‘like’, and skila ‘return’ involve allows RRG to treat dative
not as an example of quirky cases, but as one of the regular cases. This is one
of the major points of contrast between RRG and the other syntactic theories
with respect to case assignment.
(9) also accounts for the contrast between (17) and (18) with respect to case
preservation under passivization.

326 WATARU NAKAMURA
(17) Icelandic (Van Valin 1991: 150, 151)

a. Lögreglan tók Siggu fasta.
b. Sigga var tekin föst af lögreglunni.
Sigga.nom was taken fast.nom by the.police.dat
‘Sigga was arrested by the police.’
(18) Icelandic (Van Valin 1991: 152)

a. Ég hjálpaði honum.
I.nom helped him.dat
‘I helped him.’
b. Honum var hjálpaði (af mér).
him.dat was helped (by me.dat)
‘He was helped (by me).’
The patient argument Sigga receives nominative case in (17b), since it

corresponds to ‘the highest-ranking macrorole argument’ in (9a). The reason
for the preservation of dative case under passivization in (18b) is that the
non-actor argument in (18a) retains its non-macrorole status after it under-
goes passivization in (18b).6
Icelandic provides us with an opportunity to illustrate the effect of an
important typological parameter concerning the DCA (i.e. domain of
case assignment). As an initial motivation for the DCA parameter, let us
consider (19).
(19) Icelandic (Van Valin and LaPolla 1997: 578)

Jón telur mér hafa alltaf ϸótt Ólafur
John.nom believes me.dat have always thought Olaf.nom
leiðinlegur.
boring.nom
‘John believes me to have always considered Olaf boring.’
What is intriguing about (19) is that the theme argument of þykja ‘think’ in
the dependent core receives nominative case. This Icelandic example casts
doubt on the assumption shared by major syntactic theories that nomina-
tive case assignment may occur only in finite clauses and suggests the
necessity of making it an option for a language to allow (9) or (10) to apply
in each core within a clause independently.
Van Valin and LaPolla (1997: 575–581) propose to derive the nominative case
assignment in the dependent core in (19) from the DCA parameterization. The
fact that the theme argument in the dependent core receives nominative case
suggests that the DCA for Icelandic is the core, in contrast to languages such as
German, which allows no nominative-marked argument to occur in any
dependent core.7 (20) is a summary of the above discussion.
(20) DCA Parameterization

a. DCA¼Clause (e.g. English, German)
b. DCA¼Core (e.g. Icelandic, Japanese)

Case Assignment 327
The assumption that the DCA for Icelandic is the core explains the DAT-
NOM case frame of the dependent verb in (19). First, let us assume that the
dependent verb in (19) is irregular with respect to macrorole transitivity
and that the two-place verb þykja ‘think’ has the feature [MR1] in its lexical
entry. (3b) requires the only macrorole to be an undergoer, since þykja
‘think’ has no activity predicate in its LS. The AUH requires the theme
argument to be an undergoer, which leads the experiencer argument of
þykja to become a non-macrorole core argument. On the other hand, the
experiencer argument of the matrix verb telja ‘believe’ is an actor and
counts as ‘the highest-ranking macrorole argument’ in (9a). (21) shows
how the macrorole assignment proceeds in the matrix and dependent core
of (19).
(21) Macrorole assignment in (19)

Matrix core [Jón telja mér]
ACT Non-MR
Dependent core [(mér) þykja Ólafur]
(Non-MR) UND
Applying the set of case assignment rules in (9) to the dependent core in (21)
accounts for the nominative marking of the theme argument (Ólafur) in (19).
In contrast to (9), which derives accusative case systems, (10) accounts for
ergative case systems. (22a)–(22c) come from Warlpiri (Pama-Nyungan).
(22) Warlpiri (Hale 1983: 6, 13)

a. Kurdu ka parnka-mi.
child.abs aux run-npst
‘The child is running.’
b. Kurdu kapi wanti-mi.
child.abs aux fall-npst
‘The child will fall.’
c. Ngarrka-ngku ka wawirri panti-rni
man-erg aux kangaroo.abs spear-npst
‘The man is spearing the kangaroo.’
Warlpiri is a double-marking language in which the agreement clitic

system operates on a nominative-accusative basis, while independent pro-
nouns and lexical NPs are case-marked according to an ergative-absolutive
pattern. The Warlpiri case-marking pattern falls out from (10), which
assigns absolutive case to O and S arguments, while assigning ergative
case to A arguments.8
Finally, RRG extends the macrorole-based account of accusative and erga-
tive case systems to active(-stative) cross-reference/case systems, illustrated
by (23) and (24).9
(23) Acehnese (Austronesian: Durie 1985, 1987)
a. (Gopnyan) geu¼mat lôn
(3sg) 3¼hold me
‘(S)he holds me.’

328 WATARU NAKAMURA
b. Geu¼jak (gopnyan)
3¼go (3sg)
‘(S)he goes.’ (Actor)
c. Lôn ehët(¼lôn)
1sg fall(¼1sg)
‘I fall.’ (Undergoer)
(24) Western dialect of Basque (Aldai 2009: 785, 786)

a. Peru-k sagarr-a-ø jan d-u-ø.
Peru-erg apple-det-abs eaten 3sg.abs-aux-3sg.erg
‘Peru has eaten the apple.’
b. Peru-k dantzatu d-u-ø.
Peru-erg danced 3sg.abs-aux-3sg.erg
‘Peru has danced.’ (Actor)
c. Peru erori d-a.
Peru.abs fallen 3sg.abs-aux
‘Peru has fallen.’ (Undergoer)
Acehnese is a head-marking language with clitics on the verb stem that

cross-reference arguments and distinguishes between actor and under-
goer arguments in terms of where their cross-referencing clitics occur:
obligatory proclitics index actor arguments, while optional enclitics
index undergoer arguments. In contrast to Acehnese, Basque is a
double-marking language in which up to three arguments may be cross-
referenced on the auxiliary verb, while independent pronouns and lexical
NPs receive case. Aldai (2008, 2009) classifies dialects of Basque into three
major types (Western, Eastern and Central) and states that the Western
dialect marks actor and undergoer arguments with ergative and absolu-
tive case, respectively, while the Eastern dialect operates on an ergative-
absolutive basis.10
It is important to recall from Section 7.1 that RRG views verbal cross-
reference systems as being functionally equivalent to nominal case systems
(Van Valin 2013): the same set of case assignment rules applies to both cross-
reference and case systems that exhibit the split-S pattern, whether it is
realized by linear morphological slots of cross-referencing affixes/clitics or
nominal case affixes/clitics. Their unified account enables us to handle
active(-stative) case systems (most of which are head-marking) on a par with
accusative and ergative ones.
(25) and (26) are the sets of case assignment rules for the two types of
active(-stative) cross-referencing/case systems, which apply to linear morpho-
logical positions of verbal cross-referencing clitics (e.g. Acehnese) and nom-
inal case affixes (e.g. Western dialect of Basque).
(25) Case assignment rules (Accusative-active)

a. Assign nominative case to the actor argument.
b. Assign accusative case to the undergoer argument.

Case Assignment 329
(26) Case assignment rules (Ergative-active)

a. Assign ergative case to the actor argument.
b. Assign absolutive case to the undergoer argument.
To summarize, the four sets of macrorole-based case assignment rules in

(9) (accusative), (10) (ergative), (25) (accusative-active) and (26) (ergative-
active), together with the DCA parameter in (20), constitute the core part
of the RRG theory of case assignment.
7.3.2 Non-Dative Oblique Cases

We saw in the last subsection that RRG defines dative as the default case of
non-macrorole core arguments and treats it as one of the regular cases along
with the core cases. This raises the question of how RRG handles oblique
cases (or adpositions) other than dative case.
RRG defines non-dative oblique cases in terms of the LS configuration in
which the referent of the noun they mark occurs and its (non-)macrorole
status (Jolly 1991, 1993; Van Valin 1993, 2005; Van Valin and LaPolla 1997).
As a case in point, let us consider how RRG defines instrumental case, with a
focus on the instrument, implement, locatum and comitative uses of the
English instrumental preposition with.
(27) Instrument with

a. John cut the meat with a knife.
[do′ (John, [use′ (John, knife)])] CAUSE
[[do′ (knife, [cut′ (knife, meat)])] CAUSE [BECOME cut′ (meat)]]
Implement with
b. Chris wrote the letter with a pen.
do′ (Chris, [write′ (Chris, letter) ^ use′ (Chris, pen)]) & INGR exist′ (letter)
Locatum with
c. John loaded the truck with hay.
[do′ (John, Ø)] CAUSE [BECOME be-on′ (truck, hay)] (UND¼locative)
Comitative with
d. John walked to the concert with Pat. (John and Pat went to the concert.)
[do′ (John ^ Pat, [walk′ (John ^ Pat)])] & INGR be-at′ (concert, John ^ Pat)
e. Pat served wine with cheese to the guest. (Pat served wine and cheese to
the guest.)
[do′ (Pat, Ø)] & [BECOME have′ (guest, wine ^ cheese)]
Two points are worth making about (27). First, the implement with is distinct
from the instrument with, in that the instrument argument is part of a
verb’s causal chain, while the implement argument is not. Second, with
exhibits an instrumental/comitative syncretism.
Jolly (1991, 1993) makes an important observation that the instrument
with marks a potential actor that fails to function as actor, while the
locatum with marks a potential undergoer that fails to serve as undergoer
and that the comitative with marks one of the co-participants that would

330 WATARU NAKAMURA
otherwise appear as an actor or undergoer but does not. (28) is the unified
definition of the instrument, locatum and comitative uses of with.11
(28) Instrumental case rule (ICR) (Van Valin 2005: 110)

Assign instrumental case to non-MR b argument if, given two arguments, a
and b, in a logical structure, with (i) both as possible candidates for a
particular macrorole and (ii) a is equal or higher (to the left of b) on the
AUH, b is not selected as that macrorole.
The basic idea is that all of the instrument, locatum and comitative uses
of with are non-predicative: they are associated with the outcome of certain
linking options (as described in (28)), in contrast to locative prepositions (as
illustrated in (29)), which are linked to a specific LS configuration.
(29) a. John was running in the park.

LS: be-in′ (park, [do′ (John, [run′ (John)])])
b. John walked to the station.
LS: do′ (John, [walk′ (John)]) & INGR be-at′ (station, John)
The two prepositions in (29) are associated with their LSs: in introduces the
LS ‘be-in′ (. . .)’, a two-place predicate that takes the entire event (‘[do′ (John,
[run′ (John)])]’) as the second argument and situates it in a location desig-
nated by the first argument, while to introduces the LS ‘INGR be-at′ (. . .)’,
whose first argument designates the goal of the movement.
Three remarks are in order about the ICR in (28). First, (28) does not extend
to the implement with illustrated in (27b). The fact that the two-place activity
verb write has no ready-made argument slot for the implement NP a pen
disqualifies it as a possible candidate for an actor and puts the implement
use of with outside the scope of (28). Second, (28) interacts with the dative case
assignment rule in (9c) in an intricate way. For example, (27c) (repeated
below) involves a marked undergoer assignment to the locative argument,
which forces the theme argument hay to become a non-macrorole.12
(27) c. John loaded the truck with hay.
The problem is that both (9c) and (28) may apply to hay. Van Valin (2005)
argues that the ICR in (28) overrides (9c) and accounts for why hay is marked
by the instrumental preposition under the crucial assumption that both
instrumental with-phrases (as illustrated in (27a)) and locatum with-phrases
(as illustrated in (27c)) constitute a subset of non-macrorole core arguments
and that (28) applies to them.13 Finally, (28) applies to nominal cases and
adpositions that nullify the morphological distinction between instrumen-
tal and comitative, but it is important to note that their syncretism is not
common outside the Indo-European languages (Stolz et al. 2006). The fact
that the majority of languages outside the Indo-European family distinguish
the two cases (as illustrated in (30)) suggests an alternative strategy to
analyse the non-comitative and comitative uses of with separately and then

Case Assignment 331
to derive the instrumental-comitative syncretism (instead of deriving the

two uses of with from a single rule).
(30) Japanese
a. John-ga katana-de take-o kit-ta.
John-nom sword-ins bamboo-acc cut-pst
‘John cut the bamboo with a sword.’ (Instrumental)
b. John-ga Tom-to/*de eiga-o mi-ta.
John-nom Tom-COM/ins movie-acc watch-pst
‘John watched the movie with Tom.’ (Comitative)
c. John-ga kome-o niku-to/*de maze-ta.
John-nom rice-acc meat-com/ins mix-pst
‘John mixed rice with meat.’ (Comitative)
To summarize, RRG divides oblique cases into predicative and non-

predicative oblique cases: the former (e.g. ablative, allative, locative) are
associated with their LSs, while the latter (e.g. dative, instrumental) are
either assigned to a non-macrorole core argument (only when no other case
rule may apply) or are associated with the linking pattern in (28). Van Valin
(1993, 2005) and Van Valin and LaPolla (1997) lay a foundation for the
decompositional account of oblique case assignment, but they leave it as
an open question how to provide a unified account of non-dative oblique
cases/adpositions, most notably the instrumental case/adposition.
7.4 An OT Implementation
7.4.1 Regular Case Assignment

This section outlines an OT implementation of the RRG theory of case
assignment and its extension to instrumental case assignment (Nakamura
1999a, 1999b, 2015, 2021).14
OT is a constraint-based formalism with an emphasis on constraint interaction
that views a grammar as a function that maps each input to its correct structural
description (Prince and Smolensky 2004). Three fundamental principles of OT
are relevant here. First, all constraints are violable. The only requirement for a
candidate to be optimal is that it is the minimal violator in the given candidate
set. Second, a grammar resolves conflicts among constraints by ranking them in
a strict dominance hierarchy, in which each constraint has absolute priority over
all the lower-ranking constraints. Third, re-ranking of a set of individually simple
constraints yields typological variation.
Nakamura (1999a) proposes the set of OT constraints for the regular case
assignment in (31).
(31) Case assignment constraints

a. Some argument receives nominative case.
b. Undergoer arguments receive accusative case.
c. Actor arguments receive ergative case.
d. Non-macrorole core arguments receive dative case.

332 WATARU NAKAMURA
(31a)–(31d) constitute a dominance hierarchy that receives as an input a pair

of the two-tiered semantic representations of a predicate and outputs its
case frame. Unlike (9a,b) and (10a,b), (31a)–(31c) make no reference to the
ranking of actor and undergoer.
Five remarks are in order about (31a)–(31d). First, (31a) comes originally
from Jakobson (1936/1984), who analyses nominative as the default case
with no semantic content, and is required by languages such as Japanese
where every core has to have at least one nominative-marked argument.
Second, the relative ranking of (31a,d) determines whether or not a language
allows preservation of dative (or some other oblique) case under passiviza-
tion: when (31d) outranks (31a), non-macrorole core arguments cannot
receive nominative case (as illustrated by (18)), while when (31a) outranks
(31d), no case preservation under passivization is allowed.15 The latter
situation is illustrated by Japanese examples given in (32).
(32) a. John-ga Tom-ni butsukat-ta.

John-nom Tom-dat bump.into-pst
‘John bumped into Tom.’
b. Tom-ga/*ni John-ni butsuka-rare-ta.
Tom-nom/dat John-dat bump.into-pass-pst
‘Tom was bumped into by John.’
The dative-marked argument in (32a) bears nominative case under passiviza-

tion (as shown in (32b). Third, (31a) groups nominative case in accusative
systems and absolutive case in ergative systems as a single category in spite
of their distributional difference. Fourth, (31b,c) are derived from the
hypothesis that accusative and ergative case, respectively, mark undergoer
and actor arguments exclusively. Finally, various rankings of (31a)–(31c)
yield the accusative, ergative, and two types of active(-stative) case systems,
as shown in (33).16
(33) Accusative case system

a. (31d) >> (31a) >> (31b) (>> (31c)) e.g. Icelandic, German
b. (31a) >> (31d) >> (31b) (>> (31c)) e.g. Japanese, French
Ergative case system
c. (31d) >> (31a) >> (31c) (>> (31b)) e.g. Warlpiri
Accusative-active case system
d. (31d) >> (31b) >> (31a) (>> (31c)) e.g. Acehnese
Ergative-active case system
e. (31d) >> (31c) >> (31a) (>> (31b)) e.g. Western dialect of Basque
Let us see how the constraint ranking in (33a) derives the NOM-ACC and
DAT-NOM case frames in Icelandic. Tableau 7.1 shows how (33a) outputs the
NOM-ACC case frame illustrated in (12b).17
(12) b. Lögreglan tók Siggu fasta.


Case Assignment 333
Tableau 7.1 Transitive constructions in Icelandic
(31d) (31a) (31b) (31c)

NOM-NOM *! *
☞NOM-ACC *
DAT-NOM *! *
ACC-ACC *! *
Tableau 7.2 ‘Dative-subject’ constructions in Icelandic
(31d) (31a) (31b) (31c)

NOM-NOM *! * *
NOM-ACC *! *
☞DAT-NOM * *
DAT-ACC *! *
The input to the constraint hierarchy consists of an actor and undergoer

argument. The first and third candidates violate (31b), while the fourth
candidate violates (31a). In contrast, the second candidate violates the
lowest-ranking constraint alone and emerges as the winner. The winner
may violate constraints in OT, as long as the other candidates violate any
higher-ranking constraint(s).
Tableau 7.2 shows how the DAT-NOM case frame in (13a) is derived.
(13) a. Stráknum líkar slíkir bílar.

the.boy.dat likes such cars.nom
‘The boy likes such cars.’
The first and second candidates violate the top-ranking constraint (31d),
since the non-macrorole core argument receives nominative case. The fourth
candidate violates (31a), since it contains no nominative argument. The
above consideration leaves the third candidate as the winner. The third
candidate violates (31b), but it fares better than the other candidates, since
the other candidates violate either (31d) or (31a) (which is ranked higher
than (31b)).
An analogous account holds for ergative and active(-stative) case systems.
Tableau 7.3 shows how the ERG-NOM case frame is derived in (22c).
(22) c. Ngarrka-ngku ka wawirri panti-rni

man-erg aux kangaroo.nom spear-npst
‘The man is spearing the kangaroo.’
The difference between Tableau 7.1 and Tableau 7.3 comes down to the relative
ranking of (31b) and (31c): when (31b) outranks (31c), an accusative case system
emerges, while when (31c) outranks (31b), an ergative case system emerges.

334 WATARU NAKAMURA
Tableau 7.3 Transitive constructions in Warlpiri
(31d) (31a) (31c) (31b)

NOM-NOM *! *
NOM-ACC *!
☞ERG-NOM *
ERG-ACC *!
This suggests that the traditional distinction between nominative and absolu-
tive is a by-product of the relative ranking of (31b) and (31c) and that there is no
need to postulate absolutive case in addition to nominative case.
Finally, the constraint rankings in (33d) and (33e) yield accusative-active
and ergative-active case systems, respectively. When (31b) outranks (31a) as
in (33d), undergoer arguments always receive accusative marking, while
actor arguments receive nominative marking. In contrast, when (31c) out-
ranks (31a) as in (33e), actor arguments receive ergative case, while under-
goer arguments receive nominative case.
To summarize this subsection, the OT reformulation of the case assignment
rules in (9), (10), (25) and (26) in terms of the case assignment constraints in
(31) accommodates the major case frames of simple clauses in accusative,
ergative, and two types of active case systems and derives the typological
variation of case systems from re-ranking of the individually simple con-
straints that make no reference to the relative ranking of actor and undergoer.
7.4.2 Instrumental Case Assignment

Nakamura (2015, 2021) extends the domain of regular case assignment to
instrumental case within the OT-RRG framework outlined in the previous
subsection (cf. Van Valin and LaPolla 1997; Van Valin 2005, 2009). This
extension is inspired by Jakobson (1936/1984), who decomposes eight cases
in Russian in terms of three privative semantic features ([peripheral],
[directional] and [quantified]) (as partially shown in Table 7.1 below) and
defines instrumental case as having the [peripheral] feature alone (which
covers oblique core arguments and adjuncts put together), while defining
dative case as having the [peripheral] and [directional] feature.
These featural definitions suggest that dative case marks oblique argu-
ments whose referents are at the receiving end of an action (e.g. recipient
arguments of ditransitive verbs), while instrumental case serves as the
default case for adjuncts.
Three problems remain to be solved before proposing an OT-RRG
account of instrumental case assignment. First, instrumental case marks
not only a wide range of adjuncts as illustrated in (34a)–(34e), but also
non-macrorole core arguments that involve a violation of the AUH, as
illustrated in (34f).

Case Assignment 335
Table 7.1 Jakobson’s (1936/1984) featural definitions of nominative, accusative,

dative and instrumental cases
Peripheral Directional Quantified
Nominative
Accusative þ
Dative þ þ
Instrumental þ
(34) Russian (Kilby 1977: 75; Wierzbicka 1980: 23; Janda 1993: 147, 156)
a. Ivan napisal pisʹmo ručkoj.
Ivan.nom wrote letter.acc pen.ins
‘Ivan wrote the letter with a pen.’
b. Ivan pil vino litrami.
Ivan.nom drank wine.acc liters.ins
‘Ivan drank wine by the litre.’
c. Petr kivnul golovoj.
Peter.nom nodded head.ins
‘Peter nodded his head.’ (lit. ‘Peter nodded with the head.’)
d. Monax dolžen svjazatʹ usta svoi molčaniem.
monk.nom must tie.up lips.acc own.acc silence.ins
‘A monk must seal his lips with silence.’
e. Lošadʹ soseda lučše moej
horse.nom neighbour.gen better mine.gen
i siloj i krasotoj
and strength.ins and beauty.ins
‘The neighbour’s horse is better than mine both in strength and in appearance.’
f. Oni gruzili baržu drovami.
they.nom loaded barge.acc firewood.ins
‘They loaded the barge with firewood.’
In order to treat instrumental case on a par with the regular cases (i.e.
nominative, dative, accusative and ergative), it is imperative to derive all
the argument and adjunct uses of instrumental case from a single con-
straint associated with a unitary meaning. Second, the question of how
the instrumental case assignment interacts with the dative case assignment
needs to be addressed. What is at stake here is how to explain why hay in
(34f) receives instrumental marking despite being qualified to receive dative
case because of its status as a non-macrorole core argument. This is a clear
indication that it is not enough to define instrumental as the default case
for adjuncts. Finally, it remains to be shown how to derive the wide range of
interpretations an instrumental-marked noun may have from its
unitary meaning.
In order to solve the first problem, Nakamura (2015, 2021) defines instru-
mental case as the default case for everything other than macrorole argu-
ments and proposes (35b) as a constraint that represents the unitary

336 WATARU NAKAMURA
meaning of instrumental case. Nakamura goes on to revise (31d) (as shown

in (35a)) in such a way that it does not apply to non-macrorole core argu-
ments as in (34f) that involve a violation of the AUH (in this case, failure to
assign undergoer status to the lowest-ranking argument) and proposes to
rank (35a) higher than (35b) universally.18
(35) Dative and instrumental case assignment

a. Non-macrorole core arguments that do not involve a violation of the AUH
receive dative case (the underlined part is added to (31d)).
b. Case-bearing elements other than macrorole (core) arguments receive
instrumental case.
(35a,b) and their ranking recast the Jakobsonian definition of instrumen-

tal case in terms of the three-layered clausal structure and semantic
macroroles and account for why non-macrorole core arguments such as
the one in (34f) (which are beyond the scope of the decompositional
definition of instrumental case given in Table 7.1) receive instrumental
case.19
(35b) accommodates a wide range of instrumental-marked adjuncts in
Russian (and, by extension, other languages) whose interpretations are not
determined by predicate semantics, but the highly underdetermined mean-
ing of instrumental case raises the question of how to interpret each of its
adjunct uses appropriately. This third problem requires serious consider-
ation, since (35b) alone leaves an instrumental-marked noun unlinked to
any argument position in the LS and provides no clue as to the specific role it
plays in the clause.20 This amounts to a violation of the Completeness
Constraint (CC), a very general principle in RRG that governs the linking
between the syntactic and semantic representations.
(36) Completeness Constraint (Van Valin 2005: 129–130)

Jakobson (1936/1984) suggests that particular meanings of instrumental case

(e.g. instrument, implement, unit, manner, path, cause) arise through its
interaction with context, but he leaves it open what the context consists of
and how its contextual interpretation is derived.
There is no space in this chapter to provide a full account of how the
various meanings of instrumental case are derived in context (see
Nakamura (2021) for detailed discussion), but a few illustrations can
be given. First, we may follow Van Valin (2012) in deriving the meaning
of implement as in (27b) (and, by extension, (34a)) from the qualia
structure (QS) (Pustejovsky 1995) of the instrumental-marked noun
in (37a).21

Case Assignment 337
(27) b. Chris wrote the letter with a pen.
(37) a. Formal: physical-object′ (a), stationery′ (a)

Telic: do′ (b, [write′ (b, c) ^ use′ (b, a)])
Agentive: artifact′ (a)
Constitutive: . . .
b. do′ (b, [write′ (b, c)]) & INGR exist′ (c)
c. do′ (Chris, [write′ (Chris, letter) ^ use′ (Chris, pen)]) & INGR exist′ (letter)
Invoking the QS of pen (more specifically, its telic quale) for interpreting the
prepositional phrase (and the whole sentence that contains it) in (27e) leads
to introducing a new argument position for its complement noun (pen) and
thereby prevents a violation of the CC. Van Valin (2012) derives the LS of (27e)
(given in (37c)) by merging the LS in the telic quale in (37a) with the LS of the
active accomplishment use of the activity verb write in (37b). This operation
is termed co-composition (Pustejovsky 1995), an operation that derives the
meaning of a phrase compositionally from the head (in this case, the
preposition) and its argument when the latter affects the meaning of the
phrase (and the clause that contains it) beyond its role as an argument of the
head. An analogous account holds for (34a).
Second, Nakamura (2021) takes the above QS-based account of (27e) as the
first step toward identifying contextually appropriate interpretations of
instrumental-marked nouns and extends it to other uses of instrumental-
marked nouns, four of which are given in (34b)–(34e). Let us begin with (34b),
which involves a measure unit by which to quantify an indefinite amount of
some object, while engaging in some activity involving it (in this case,
drinking wine). We may take the instrumental-marked noun litrami ‘litres’
as an adjunct, since it is not subcategorized by the verb. The underspecified
meaning of instrumental case requires the unit noun litr ‘litre’ to be linked
to an argument position of an event involved by its QS. I propose that the
unit interpretation of litrami is derived from its telic qualia given in (38a): litr
‘litre’ refers to a unit that is used to measure an amount of liquid.
(38) a. Qualia structure of litr ‘litre’

TELIC: do′ (x, [use′ (x, y)]) PURP do′ (x, [measure′ (x, z)])
b. LS of pitʹ: do′ (x, [drink′ (x, z)])
c. LS of (34b): do′ (Ivan, [drink′ (Ivan, wine)] ^ [use′ (Ivan, litre)
PURP measure′ (Ivan, wine)])
The CC requires us to interpret litrami ‘litres’ as a participant of the event

involved by its telic quale. Merging the LS of the activity verb pitʹ ‘drink’ in
(34b) with the LS in the telic quale of litr in (38a) yields the LS of (34b)
in (38c).
Third, the instrumental-marked noun in (34c) fulfils two functions simul-
taneously. First, the body-part noun golovoj ‘head’ serves as an adjunct to the
main predicate and restricts the verb’s meaning by identifying the body part
that is directly involved in the event rather than designating a target of

338 WATARU NAKAMURA
Peter’s action (Wierzbicka 1980). Second, the body-part noun forms a posses-
sive relation with the subject argument: its constitutive quale (‘have.as.part′
(x:human, y)’) allows the hearer to associate Petr and golovoj with the posses-
sor (‘x’) and possessum (‘y’), respectively. Another point to note in this
connection is that (34c) encodes the possessive relation in terms of predica-
tion (and not in terms of reference, as illustrated in (39)) despite the fact that
the possessive predicate remains covert.
(39) golova Petra

head.nom Peter.gen
‘Peter’s head’
(40) a. Main predicate: SEML do′ (Peter, [nod′ (Peter)])

b. Covert possessive predicate: have.as.part′ (Peter, head)
The main predicate kivnutʹ ‘nod’ and the covert possessive predicate share
the subject argument Petr, but these predicates do not form a nuclear
juncture, since Petr participates in the two events in (40a,b) independently.22
This suggests that (34c) embodies a non-subordinate core juncture, in which
each of the predicates has its own core.
The macrorole assignment proceeds in the two cores independently. The
main predicate kivnutʹ ‘nod’ is a single-macrorole verb. Petr functions as
undergoer when it counts as the first argument of the covert possessive
predicate in (40b) (since it has no activity predicate do′ in its LS), while Petr
serves as actor when it counts as the effector argument of the main predi-
cate. The possessum argument golova ‘head’ becomes a non-macrorole core
argument and receives instrumental case from (35b), since the possessor
argument Petr bears the undergoer status in violation of the AUH.23
In contrast to (34a)–(34c), the instrumental-marked noun in (34d) (molčanie
‘silence’) has no (default) value for its constitutive, telic or agentive quale: it
is not conventionally associated with any particular material, purpose,
function or origin. This yields the consequence that we need to understand
the function of molčanie with no reference to its QS. Since molčanie is neither
an argument of the predicate nor part of the predicate, we have to conclude
by a process of elimination that this instrumental-marked noun serves as a
modifier: it modifies the manner of the action denoted by the verb svjazatʹ
‘tie up.’
Finally, let us consider how the meaning of domain restriction is derived
in (34e). The first point to note here is that both sila ‘strength’ and krasota
‘beauty, appearance’ are part of the formal qualia of lošadʹ ‘horse.’ The fact
that these attributes constitute part of the attributes of a horse explains why
the two instrumental-marked abstract nouns serve to relativize the evalu-
ation of the neighbour’s horse by the speaker and receive the interpretation
of domain restriction. The fact that the two abstract nouns are not subcat-
egorized by lučše suggests that they function as a modifier of the predicate
and receive instrumental case from (35b) as dominated by (35a) (since they
are not syntactic arguments).

Case Assignment 339
To summarize, this section has shown that the OT reformulation of the

case assignment rules in (9) (accusative), (10) (ergative), (25) (accusative-
active) and (26) (ergative-active) clarify what these case systems share and
where they diverge and has provided a brief summary of how to extend the
OT-RRG account of the regular cases to instrumental case.
7.5 An OT-RRG Account of Case Syncretism
One of the major principles of the RRG theory of case assignment (Van Valin
1993, 2005; Van Valin and LaPolla 1997) is that it does not distinguish syntactic
cases from morphological cases, under the assumption that each case receives
its unique morphological realization (see Section 7.1). However, this assump-
tion is called into question by case syncretism phenomena, where a single case
morpheme realizes more than one syntactic case. This section outlines the OT-
RRG account of case syncretism (based mainly on Nakamura (2011)), with a
particular focus on those instances of syncretism in which a single case
morpheme realizes more than one regular case other than instrumental case
(i.e. nominative, accusative, ergative, dative and genitive).
The rest of this section is organized as follows. First, I will introduce the
case hierarchy (CH) (Silverstein 1980/1993; cf. Blake 2001; Caha 2013) and
will summarize what the CH has to say about the typological variation of
case syncretism among the regular cases. Second, I will illustrate the repre-
sentative types of syncretism predicted by the CH. Finally, I will show how
and to what extent these instances of case syncretism are derivable in terms
of OT.
7.5.1 The Case Hierarchy

The CH comprises two parts: the upper part represents implicational rela-
tions among propositional and adnominal case morphemes, while the lower
part represents implicational relations among adverbial and propositional
case morphemes.
(41) Case Hierarchy (Silverstein 1976, 1980/1993)24

a. Propositional/Adnominal: Nom/Abs: Dat1 ⟵ {Acc, Erg} ⟵ Gen
b. Adverbial/Propositional: Dat2 ⟵ Instr, Loc. . .
(41a) states that if a language has a distinct morpheme for genitive, it has a
distinct morpheme for accusative and/or ergative, that if a language has a
distinct morpheme for accusative and/or ergative, it has a distinct mor-
pheme for dative and nominative/absolutive, while (41b) states that if a
language has a distinct morpheme for representing instrumental and/or
locative, it has a distinct morpheme for dative.
Five remarks are in order about (41). First, (41a) states that nominative/
absolutive and dative constitute the minimal case-marking system and that

340 WATARU NAKAMURA
three/four/five-way case-marking systems arise as a gradual elaboration of this

fundamental contrast.25 Second, (41a) treats nominative and absolutive as
different manifestations of the same case (Section 7.4.1). Third, some languages
(e.g. Estonian) have no case morpheme termed ‘dative’, but they have some
other oblique case morpheme that behaves like it (Matsumura 1994, 1996; see
Section 7.5.2). Fourth, a genitive case expresses a possessive relation as distin-
guished from an attributive modification. This means that attributive markers
are not analysed as genitive.26 Finally, (41a) and split accusativity and ergativity
determine the possible range of case syncretism among the regular cases. (42) is
a list of the patterns of case syncretism predicted by (41a).
(42) Variation of case syncretism

a. Gen¼Erg¼Acc¼Dat, Nom (e.g. Yagnob)
b. Gen¼Erg¼Dat, Nom (e.g. Kabardian)
c. Gen¼Acc¼Dat, Nom (e.g. Palauan)
d. Gen¼Dat, Acc, Nom (e.g. Old Persian)
e. Gen¼Dat, Erg, Nom (e.g. Djaru)
f. Gen¼Acc, Ade(Dat), Nom (e.g. Estonian)
g. Gen¼Erg, Dat, Nom (e.g. Inuktitut)
h. Gen¼Erg, Acc¼Dat, Nom (e.g. Tagalog)
i. (Gen), Erg, Acc¼Dat, Nom (e.g. Hindi)
j. Acc¼Dat, Nom (e.g. Spanish)
7.5.2 Patterns of Case Syncretism

(42a)–(42c) are two-way case-marking systems illustrated by Yagnob (Iranian),
Kabardian (Northwest Caucasian), and Palauan (Austronesian). They exhibit
an extensive syncretism that applies to all non-nominative case morphemes
available. (43a)–(43f) come from Kabardian.
(43) Kabardian (Colarusso 1992: 167; Smith 1996: 108, 111, 113)
a. ɬ’ -m
e š -r e f ́ z -m
e e j r jtáhs.
e e
˙
man-obl horse-nom woman-obl (nom.3).io.act.gave
‘The man gave the horse to the woman.’
b. ɬ’ -m
e š -r e j -w h’áhs.
e e
˙
man-obl horse-nom (nom.3)-act-killed
‘The man killed the horse.’
c. ha-r žás -m e mabáhna.
˘ ˙
dog-nom night-obl (nom.3)-bark
‘The dog barks at night.’
d. ɬ’ -r
e f ́ z -m
e e náxra nax’’ ́ zs. e
˙˙
man-nom woman-obl older (nom.3)-is
‘The man is older than the woman.’
e. máz -m e jahh.
˘
forest-obl act.(nom.3).carry
‘They carry it to the forest.’
f. ha-m Ø-y -pa-re
˘
dog-obl 3-poss-nose-nom
‘the dog’s nose’

Case Assignment 341
The nominative case suffix -r marks S and O arguments, while the oblique
case suffix -m marks A arguments, a wide range of oblique core arguments
and adjuncts, and adnominal possessors. In contrast to the nominative case
suffix, which may appear only once per clause, the oblique case suffix may
appear multiple times within a clause. The fact that Kabardian distinguishes
A arguments from S and O arguments in terms of case marking suggests
that its case system (as well as its cross-referencing system) involves an
ergative alignment.
In contrast to Kabardian, Palauan involves an accusative case alignment.
(44a)–(44f) show that it uses the oblique preposition er to mark non-
macrorole core arguments, human and/or specific-and-singular
O arguments in imperfective clauses, adjuncts and adnominal possessors,
while leaving S and A arguments unmarked.27
(44) Palauan (Josephs 1975: 235; Georgopoulos 1991: 26, 27, 29)
a. ak-mo er a katsudo.
r.1sg-go obl movies
‘I am going to the movies.’
b. ng-kiltmekl-ii a ulaol a Peter.
r.3sg-clean-3sg floor Peter
‘Peter cleaned the floor.’
c. ng-diak ku-nguiu er a hong.
neg irr.1sg-read obl book
‘I am not reading the book.’
d. ng-mo er a ngebard er a klukuk.
r.3sg-go obl west obl tomorrow
‘She is going to America tomorrow.’
e. A Romana a omeka er a rengalek
Romana feed obl children
er a kukau.
obl taro
‘Romana is feeding the children the taro.’
f. ak-uleldanges er a resensei er ngak
r.1sg-honor.ipfv obl teachers obl me
‘I respected my teachers.’
Tables 7.2(a,b) represent the extensive case syncretism in Kabardian and

Palauan.
Yagnob (Iranian) exhibits an even broader range of syncretism than Kabar-
dian and Palauan: non-macrorole core arguments, A arguments in ergative
constructions, definite O arguments in accusative constructions, and
adnominal possessors receive the same oblique case marker -i (Comrie
1981: 169–170).
(42d)–(42i) illustrate three-way case-marking systems with a genitive-
dative, genitive-ergative, genitive-accusative, or dative-accusative syncre-
tism. Let us begin with the genitive-dative syncretism in (42d,e). (45a)–(45c)
come from Djaru (Pama-Nyungan).

342 WATARU NAKAMURA
Table 7.2(a) Case syncretism in Kabardian
Case Case marking
NOM -r
ERG -m (oblique)
GEN
DAT
INSTR, LOC . . .
Table 7.2(b) Case syncretism in Palauan
Case Case marking
NOM ø
ACC er (oblique)
GEN
DAT
INSTR, LOC . . .
(45) Djaru (Tsunoda 1981: 110, 115, 124)

a. ŋaniŋa guɲar guɳga ɲir-a waɽulu-la.
1sg.dat dog dead stay-pst night-loc
‘My dog died last night.’
b. ŋumbir-u ŋa-la maŋari jambagina-wu juŋ-an.
woman-erg child-3sg.dat food child-dat give-prs
‘A woman gives food to a child.’
c. ŋaᶁu-ŋgu ŋa-ɳa-ŋgu guju man-i
1sg-erg c-1sg.nom-3sg.dat meat take/get-pst
ɲunuŋa.
2sg.dat
‘I obtained meat for you.’ (or ‘I took/stole your meat.’)
(45a,b) indicate that the same case morpheme is used to mark recipient
arguments of ditransitive verbs and adnominal possessors. This genitive-
dative syncretism renders (45c) ambiguous and allows the second-person
dative pronoun in (45c) to be construed as a benefactive or possessor (Tsu-
noda 1981: 110).
Likewise, Old Persian (Indo-Iranian) exhibits the same syncretism in its
accusative case system (Benvenuto and Pompeo 2015). Romanian (Romance)
provides another illustration of genitive-dative syncretism.
(46) Romanian (Cornilescu 2000: 91; Hill 2013: 140)

a. cartea profesorului
book.def teacher.def.obl
‘the teacher’s book’
b. Copiii n-au raspuns
~ deloc bine profesoarei.
children neg-have responded at.all well teacher.def.obl
‘Children have not responded at all well to the teacher.’

Case Assignment 343
c. Ion li-a admirat pe prietenuli lui.

Ion him-has admired acc friend.def his
‘Ion admired his friend.’
Table 7.3 Case syncretism in Romanian
Case marking
Case Inflection Preposition
NOM Nominative
ACC pe
GEN Oblique
DAT
Romanian has a two-way inflectional case-marking system for nouns, while

using a differential object marker pe usually accompanied by clitic doubling
(as illustrated in (46c)) to mark those undergoer arguments of transitive
verbs that are relatively high on the definiteness and animacy scales (see
Mardale (2008) for discussion of how the two scales are combined). Table 7.3
is a summary of the hybrid case-marking system of Romanian.28
The Romanian case-marking system involves split-accusativity triggered
by definiteness and animacy in addition to the genitive-dative syncretism.
The genitive-dative syncretism is also attested in Macedonian, Bulgarian,
Albanian and Modern Greek and has been recognized as one of the Balkan
Sprachbund properties (Tomić 2004: 12–15).
Halkomelem (Salish) is a double-marking language that exhibits a two-way
case-marking system for nouns, while possessing a split-ergative agreement
system:29 it involves a genitive-dative syncretism (using the same oblique
case suffix to mark oblique nouns and proper-noun adnominal possessors)
and makes no formal distinction among A, O and S arguments, all of which
receive nominative (‘straight’) case (Gerdts 1988).
The genitive-accusative syncretism is attested in Northern Sámi,
Estonian and Finnish (Finno-Ugric). As a point of departure, let us consider
the case-marking system in Northern Sámi. The genitive/accusative case
morpheme in Northern Sámi marks (i) complement nouns of postpositions
and prepositions, (ii) subject arguments of non-finite verbs, and (iii) nouns
that designate a standard of comparison in comparative constructions, in
addition to adnominal possessors and undergoer arguments of transitive
verbs.30
(47) Northern Sámi (Valijärvi and Kahn 2017: 52, 53)

a. Máret orru vánhemiid luhtte.
Máret lives parents.gen.pl with
‘Máret lives with her parents.’
b. Son ođii Máhte humadettiin.
3sg.nom slept Máhtte.gen.sg talk.ger
‘S/he slept while Máhtte was talking.’

344 WATARU NAKAMURA
c. Biila lea skohtera ođđaseabbo.

car.nom.sg is snowmobile.gen.sg newer
‘The car is newer than the snowmobile.’
The genitive-accusative syncretism holds across the board in Northern Sámi.

The semantic and syntactic diversity of genitive/accusative-marked nouns
indicates that the genitive/accusative case is assigned to nouns that would
remain caseless otherwise.
Estonian represents a more complex situation than Northern Sámi, in that
the genitive-accusative syncretism holds only in the singular, and the geni-
tive/accusative case alternates with the partitive case when it occurs on
undergoer arguments of transitive verbs. When the genitive/accusative case
alternates with the partitive case, it serves as an indicator of aspectual
boundedness, as demonstrated by the contrast between (48b) and (48c).31
(48d) indicates that when transitive constructions with plural undergoer
arguments denote a telic event, they receive the nominative case (instead of
the genitive/accusative case).
(48) Estonian (Miljan 2008: 2, 177)

a. maja aknad
house.gen.sg window.nom.pl
‘windows of the house’
b. Raul ehitas suvila.
Raul.nom build.pst.3sg cottage.gen.sg
‘Raul built a cottage.’ (completed)
c. Raul ehitas suvila-t.
Raul.nom build.pst.3sg cottage-prt.sg
‘Raul was building a cottage.’ (incomplete)
d. Kass sõi hiire/hiired ära.
cat.nom.sg eat.pst.3sg mouse.gen.sg/mouse.nom.pl up
‘The cat ate the mouse/the mice.’
Table 7.4 shows how undergoer arguments of transitive verbs are case-
marked in Estonian.
Furthermore, Finnish syncretizes genitive and accusative in the singular
declension, exhibits an aspectually conditioned case alternation on under-
goer arguments of transitive verbs (which is analogous to the one shown in
Table 7.4), and allows the genitive/accusative case to mark a wide variety of
nouns, including complement nouns of many postpositions and subject
Table 7.4 Case marking of O arguments in Estonian
Number Telicity Case marking
O arguments singular bounded Genitive/Accusative

unbounded Partitive
plural bounded Nominative

unbounded Partitive

Case Assignment 345
arguments of non-finite verbs, but Finnish is in contrast to Estonian, in that

it has distinct accusative case forms for first-person pronouns, second-person
pronouns, and third-person animate pronouns and animate interrogative
pronouns.
A genitive-ergative syncretism is attested in Tagalog (Austronesian),
Inuktitut (Eskimo-Aleut) and Mayan languages. (49a)–(49c) come from
Inuktitut.
(49) Inuktitut (Johns 1992: 58, 59, 69)

a. arna-up angut kuni-ga-a.
woman-rel man.abs kiss-pass.ptcp-3sg/3sg
‘John is stabbing/stabbed the seal.’
b. angut ani-juq.
man.abs go.out-intr.ptcp.3sg
‘The man went out.’
c. anguti-up qimmi-a
man-rel dog-3sg
‘the man’s dog’
The case morpheme that marks adnominal possessors and actor arguments
of transitive verbs in the Eskimo languages has been termed ‘relative case’ in
Eskimo linguistics. This and similar instances of the genitive-ergative syn-
cretism have repeatedly been argued to have arisen from (a reanalysis of )
nominalized constructions (Trask 1979; Bricker 1981; Starosta et al. 1982;
Johns 1992; Kaufman 2009; Coon 2009, 2013). The consensus in the literature
cited here is that the syntactic parallelism between the possessive and
transitive constructions goes a long way toward accounting for the
genitive-ergative syncretism.
Another example of the genitive-ergative syncretism comes from Tagalog,
which displays a symmetrical voice system with verbs bearing voice morph-
ology to indicate the semantic role of a nominative-marked argument, as
illustrated by (50a)–(50h).
(50) Tagalog (Kroeger 1993: 13, 14, 23, 32, 50; Sabbagh 2016: 658)
a. Pinutol ng¼magsasaka ang¼sungay ng¼kalabaw.
pfv.pv.cut gen¼farmer nom¼horn gen¼buffalo
‘The farmer cut off the buffalo’s horn.’ (Patient voice)
b. Ibinigay lahat ng¼mga¼guro sa¼mga¼bata
pfv.pv.give all gen¼pl¼teacher dat¼pl¼child
ang¼pera.
nom¼money
‘The teachers gave all the money to the children.’ (Patient voice)
c. Bumili ang¼lalake ng¼isda sa¼tindahan.
pfv.av.buy nom¼man gen¼fish dat¼store
‘The man bought fish at the store.’ (Actor voice)
d. Binili ng¼lalake ang¼isda sa¼tindahan.
pfv.pv.buy gen¼man nom¼fish dat¼store
‘The man bought the fish at the store.’ (Patient voice)

346 WATARU NAKAMURA
e. Ibinili ng¼lalake ng¼isda ang¼bata.

pfv.bv.buy gen¼man gen¼fish nom¼child
‘The man bought fish for the child.’ (Benefactive voice)
f. Tanging pumansin kay¼Elias si¼Maria Clara.
only pfv.av.notice dat¼Elias nom¼Maria Clara
‘Only Maria Clara noticed Elias.’ (Actor voice)
g. Kaaalis pa lamang ni¼Pedro nang
rec.pfv.leave yet only gen¼Pedro adv
dumating ako.
pfv.av.arrive 1sg.nom
‘Pedro had just left when I arrived.’ (Voiceless)
h. Kapangunguha pa lamang ng¼bata ng¼mga¼mangga.
rec.pfv.gather yet only gen¼child gen¼pl¼mango
‘The child has just gathered some/the mangoes.’ (Voiceless)
Four comments are in order about (50a)–(50h). First, (50a,c,d,e) show that
the genitive case marker ng/ni encodes not only adnominal possessors,
but also transitive actor arguments in non-actor (i.e. patient, benefactive,
locative or instrumental) voice constructions and transitive undergoer
arguments in non-patient voice constructions. Second, (50c) and (50d)
show that nominative-marked undergoer arguments of transitive verbs
in non-actor voice constructions are construed as specific (or generic),
while genitive-marked undergoer arguments of transitive verbs in actor
voice constructions are typically construed as non-specific.32 Third, Taga-
log requires undergoer arguments of transitive verbs to receive dative
case when they are pronouns or proper names that refer to animate
entities (as illustrated in (50f)).33 We may interpret the oblique marking
of Elias in (50f) as an instance of differential object marking, under the
assumption that sa syncretizes dative and accusative. Finally, Tagalog has
a voiceless construction with no overt dependency between the verb
inflected for recent-perfective aspect and one of its arguments: the verb
bears no voice morphology and neither of its arguments receives nom-
inative marking (Kroeger 1993: 50).34 (50g,h) show that single arguments
of intransitive verbs and two major arguments of transitive verbs receive
genitive case in recent-perfective constructions.35 This suggests that ng/ni
serves as the default case marker of core arguments (cf. Foley and Van
Valin 1984: 389).
Finally, the accusative-dative syncretism is found in a number of Indo-
Aryan languages (e.g. Hindi, Punjabi, Kashmiri) and Romance languages (e.g.
Spanish, Catalan, Southern dialects of Italian). For example, Hindi uses the
case clitic -ko to encode animate and/or specific undergoer arguments of
transitive verbs and recipient arguments of ditransitive verbs.
(51) Hindi (Mohanan 1994: 59, 80)

a. ilaa-ne haar-ko uthaayaa.
_
Ila-erg necklace-acc lift.pfv
‘Ila lifted the/*a necklace.’

Case Assignment 347
b. ilaa-ne bacce-ko/*baccaa uthaayaa.

_
Ila-erg child-acc/child.nom lift.pfv
‘Ila lifted a/the child.’
c. niinaa-ne bacce-ko kitaab dii.
Nina-erg child-dat book.nom give.pfv
‘Nina gave the child a book.’
Likewise, Spanish uses the preposition a to mark not only recipient argu-
ments of ditransitive verbs and goal arguments of intransitive motion verbs,
but also undergoer arguments of transitive verbs that are typically both
human and specific.
(52) Spanish (Zagona 2001: 13, 14)

a. En el mercado vi *(a) los vecinos.
at the market saw-1sg (acc) the neighbours
‘At the market (I) saw the neighbours.’
b. En el escritorio vi (*a) los papeles.
on the desk saw-1sg (acc) the papers
‘On the desk (I) saw the papers.’
c. Juan lei mandó un paquete a Joséi.
Juan dat sent-3sg a package dat José
‘Juan sent a package to José.’
Since Spanish has no genitive case morpheme in both nouns and pronouns,
it exhibits the two-way case-marking system (consisting of nominative and
dative) for nouns and local (first and second) person pronouns and the three-
way case-marking system (consisting of nominative, dative and accusative)
for third-person pronouns.36
7.5.3 ‘Dative’ Case in Northern Sámi, Estonian, Finnish and

Tolmači Karelian
Before considering how to derive the variation of case syncretism, a digres-
sion is in order about whether or not the case-marking systems in Northern
Sámi, Estonian, Finnish and Tolmači Karelian fall within the scope of the
CH, since they seem to have no dedicated ‘dative’ case morpheme despite
having an accusative or genitive/accusative case morpheme.
Let us begin with Northern Sámi.
(53) Northern Sámi (Aikio 2009; Valijärvi and Kahn 2017: 54–56)
a. Bussá čohkká stuolus.
cat.nom sit.prs.3sg chair.loc
‘The cat is sitting on a chair’ (Static location)
b. Bussá njuike stuolus láhttái.
cat.nom jump.prs.3sg chair.loc floor.ill
‘The cat jumps from the chair onto the floor.’ (Source)
c. Nieiddas lea ođđa irgi.
girl.loc be.prs.3sg new boyfriend.nom
‘The girl has a new boyfriend.’ (Possessor)

348 WATARU NAKAMURA
d. Mun lean ožžon e-boastta

1sg.nom be.prs.1sg receive.pst.ptcp e-mail.gen
Ristenis.
Risten.loc
‘I have received an e-mail from Risten.’ (Sender)
e. Mun adden beatnagii dávtti.
1sg.nom give.pst.1sg dog.ill bone.gen
‘I gave a bone to the dog.’ (Receiver)
f. Máhtes oaivi bávččasta.
Matthew.loc head.nom ache.prs.3sg
‘Matthew has a headache.’ (Experiencer)
g. Máhtes dállu bulii.
Matthew.loc house.nom burn.pst.3sg
‘Matthew’s house burned down on him.’ (Experiencer)
h. Máhtes gahčai lássa láhttái.
Matthew.loc fall.pst.3sg glass.nom floor.ill
‘Matthew dropped a glass on the floor.’ (Low degree of agentivity)
(53a)–(53e) show that the locative case conflates the distinction between
location and source in both the concrete local and possessive domains,
while the illative case marks goal arguments of motion verbs and recipient
arguments of ditransitive verbs. (53f,g) show that the locative case may
mark animate experiencer arguments and demonstrate that the locative
case has detached itself from the concrete locative system and has become
able to mark a non-spatial (as well as spatial) participant of an event.37
(53h) illustrates a further step taken by the locative case to move out of the
physical local domain, since it construes the action taken by Matthew as
involuntary or beyond his control (Aikio 2009).38 The fact that the locative
case conflates the location/source distinction and covers the dative domain
(Seržant 2015) except for recipient arguments and has been grammatica-
lized to be able to form a paradigmatic contrast with the nominative case
and to denote the low degree of agentivity suggests that the locative case
in Northern Sámi is abstracted from concrete locative domains to repre-
sent pure obliqueness, which the dative case in other languages is sup-
posed to represent.39
Second, the adessive case (one of the external local cases) in Estonian
encodes possessive and other non-spatial relations much more frequently
than spatial ones (Matsumura 1994); it marks possessor arguments, experi-
encer arguments, and causee arguments of causative verbs, and ‘logical
subjects’ of impersonal verbs, as illustrated by (54a)–(54i).40
(54) Estonian (Matsumura 1994: 225, 230, 231; Lindström and Vihman 2017: 2, 10)
a. Raamat on laua-l.
book.nom be.prs.3sg desk-ade
‘The book is on the desk.’ (Location)
b. Mu-l on uus auto.
1sg-ade be.prs.3sg new.nom car.nom
‘I have a new car.’ (Possessor)

Case Assignment 349
c. Ta-l sur-i laps.

3sg-ade die-pst.3sg child.nom
‘Her child died.’ (Deprived possessor)
d. Ta-l on soov kodus olla.
3sg-ade be.prs.3sg wish.nom at.home be.inf
‘She has a wish to be at home.’ (Possessor with an infinitival complement)
e. Mu-l on tema-st kahju.
1sg-ade be.prs.3sg s/he-ela sorry
‘I feel sorry for him/her.’ (Experiencer)
f. Mu-l/lle on vaja uut arvuti-t.
1sg-ade/all be.prs.3sg need new.prt computer-prt
‘I need a new computer.’ (Modal experience with a nominal complement)
g. Mu-l on vaja koju minna.
1sg-ade be.prs.3sg need home.ill go.inf
‘I need to go home.’ (Modal experiencer with an infinitival complement)
h. Jaan lase-b ta-l vene keele-s
Jaan.nom let-prs.3sg 3sg-ade Russian language-ine
vasta-ta.
answer-inf
‘Jaan makes/lets her answer in Russian.’ (Causee)
i. Ta-l õnnestu-s ülikooli astu-da.
3sg-ade succeed-pst.3sg university.ill enter-inf
‘She succeeded in entering the university.’
The fact that the adessive case not only covers part of the dative domain (i.e.
possessor and experiencer arguments) but also causee arguments of transi-
tive verbs and ‘logical subjects’ of impersonal verbs (e.g. (54i)) leads Matsu-
mura (1994) to define the adessive as a functional equivalent to dative
(termed ‘adessive-dative’). The allative case also marks some experiencer
arguments as well as recipient arguments of ditransitive verbs, as illustrated
in (55a,b).
(55) Estonian (Lindström and Vihman 2017: 10)

a. Ta and-i-s mu-lle medali
3sg.nom give-pst-3sg 1sg-all medal.gen
‘S/he gave me a medal.’ (Recipient)
b. Mu-lle meeldi-b su uus müts.
1sg-all like-3sg 2sg.gen new.nom hat.nom
‘I like your new hat.’ (Experiencer)
However, the crucial difference between the adessive and allative is that
experiencer arguments require or prefer the adessive marking when
accompanied by an infinitival complement (as illustrated by (54g)) (Lind-
ström and Vihman 2017).41 (54h,i) represents a further departure from
the locative use in (54a), since the adessive case in (54h,i) only marks the
highest-ranking direct core argument (‘logical subject’) of the embedded
infinitival verb. We may take these facts as an indication that the ades-
sive case in Estonian has been grammaticalized to be able to behave as

350 WATARU NAKAMURA
the ‘pure’ oblique marker abstracted from concrete locative relations

(Matsumura 1994).
The wide range of functions encoded by the adessive case in Estonian
suggests that, like the locative case in Northern Sámi, the adessive case is
more prone to grammaticalization than the allative case and hence behaves
as the least-marked oblique case morpheme in Estonian.
Third, Finnish partitions the dative domain in such a way that possessor
and experiencer arguments receive the genitive or adessive (or some other
local) case, while recipient arguments of ditransitive verbs receive the alla-
tive case, as illustrated by (56a)–(56j).42
(56) Finnish (Sulkala and Karalainen 1992: 295; Huumo 1996: 81, 82; Karlsson 1999:
96; Kiparsky 2001: 356; Metslang and Erelt 2006: 263; Mazzitelli 2017: 28)
a. Poja-lla/*-n ol-i kirje.
boy-ade/gen be-pst.3sg letter.nom
‘The boy had a letter.’ (Possessor)
b. Poja-lta/*-n katos-i kolikko.
boy-abl/gen disappear-pst.3sg coin.nom
‘The boy lost a coin.’ (Deprived possessor)
c. Minu-n/lla ol-i nälkä.
1sg-gen/ade be-pst.3sg hunger.nom
‘I was hungry’ (Experiencer)
d. Minu-n/lla on ikävä hän-tä.
1sg-gen/ade be.prs.3sg sorry.nom he-prt
‘I feel sorry for him.’ (Experiencer)
e. Mies-ten on pakko poistua.
man-gen.pl be.prs.3sg obligation.nom leave.inf
‘The men have to leave.’ (Modal experiencer)
f. Mauno-n ol-i hauska pääs-tä kotiin.
Mauno-gen be-pst.3sg pleasant get-inf home.ill
‘It was nice for Mauno to get home.’
g. Vireni-n onnistui voittaa.
Viren-gen succeed.pst.3sg win.inf
‘Viren succeeded in winning.’
h. Anno-i-n sinu-n/Mati-n näh-dä karhu-n.
let-pst-1sg 2sg-gen/Matti-gen see-inf bear-gen
‘I let you/Matti see a/the bear.’
i. Pauli kirjoituttaa Harri-lla kirje-en.
Pauli.nom write.caus.prs.3sg Harri-ade letter-gen
‘Pauli makes Harri write a letter.’ (Causee)
j. Anno-i-n hei-lle karhu-n.
give-pst-1sg 3pl-all bear-gen
‘I gave them a bear.’ (Recipient)
The uses of the genitive case in (56c)–(56h) (termed ‘dative-genitive’ in the

literature) are distinguished from those uses of it that mark adnominal
possessors and undergoer arguments of transitive verbs. Comparing (56a)–
(56i) with (54a)–(54i) reveals that the genitive case in predicative possessive

Case Assignment 351
Table 7.5 Dative domain in Northern Sámi, Estonian, Finnish and Tolmači Karelian
Northern Sámi Estonian Finnish Tolmači Karelian
Possessor LOC ADE ADE, ABL ADE

Experiencer LOC ADE/ALL GEN/ADE ADE
Causee ADE GEN/ADE
Recipient ILL ALL ALL ADE
constructions has been replaced by the external local cases (including the
adessive case) and those uses of the genitive case that mark experiencer
arguments are competing with the adessive case (as in (56c,d)). (56h,i) show
that causees of transitive verbs receive genitive or adessive case, depending
on whether they occur in periphrastic or morphological causative construc-
tions.43 What is peculiar about Finnish is that the adessive case has been
replacing the ‘dative-genitive’ case, in contrast to Northern Sámi and Esto-
nian, which, respectively, use the locative and adessive case as some kind of
‘pure’ oblique marker.44
Finally, Tolmači Karelian, a dialect of Karelian (Finnic), uses the adessive
case to mark not only possessor and experiencer arguments but also recipi-
ent arguments of ditransitive verbs, due to the loss of the allative case
morpheme (which formerly marked goal and recipient arguments) (Oranen
2019). Oranen chooses to retain the traditional term ‘adessive’ (instead of
‘dative’) for etymological and historical reasons, but notes that the adessive
case in Tolmači Karelian is comparable to the dative case in Russian.45
To summarize, Northern Sámi, Estonian and Finnish use a different
case morpheme for recipient arguments of ditransitive verbs than the
one used for the rest of the dative domain (in contrast to Tolmači Kare-
lian, which uses the adessive case for the entire dative domain), but the
fact that the locative case (Northern Sámi), the adessive case (Estonian)
and the genitive or adessive case (Finnish) behave, to varying degrees, as a
‘pure’ oblique marker abstracted from concrete locative domains in a
paradigmatic contrast with the core cases suggests that the CH holds true
even in languages with no dedicated ‘dative’ case morpheme. Table 7.5
summarizes the data from the four languages that are discussed in
this subsection.
7.5.4 Deriving Case Syncretism

In order to derive the patterns of case syncretism in (42), Nakamura (2011)
introduces the distinction between syntactic and morphological cases into
RRG and derives their multiple correspondence from an OT-style
constraint hierarchy.
There are three steps in deriving the patterns of case syncretism in (42).
The first step is to recast the case assignment constraints in (31) and another

352 WATARU NAKAMURA
one that assigns genitive case to adnominal possessors as constraints for

syntactic case assignment.
The second step is to propose two types of constraints that determine the
correspondence between syntactic cases and morphological cases: marked-
ness and faithfulness constraints. The markedness constraints are derived
from the CH in (41a).
(41) Case Hierarchy (Silverstein 1976, 1980/1993)

a. Propositional/Adnominal: Nom/Abs: Dat1 ⟵ {Acc, Erg} ⟵ Gen
(41a) ranks the propositional and adnominal case morphemes according to

their morphological markedness. ‘{Acc, Erg}’ indicates that either of the two
case morphemes may be missing and that there is no inherent markedness
relation between the two. This means that ‘*Erg’ may outrank ‘*Acc’ or the
other way around when both of them are available. From (41a), we may derive
a fixed hierarchy of markedness constraints in (57a), which compete against
two faithfulness constraints in (57b,c), ‘MAX [Case]’ and ‘IDENT [Case]’ (‘*Acc,
*Erg’ in (57)–(59) indicates that ‘*Acc’ and ‘*Erg’ are equally ranked).46
(57) a. *Gen >> {*Erg, *Acc} >> *Dat

b. MAX [Case]
c. IDENT [Case] (i.e. IDENT [Gen, Acc, Dat] & IDENT [Gen, Erg, Dat])
‘MAX [Case]’ requires each case feature value in the input (syntactic case) to
be realized by some case morpheme (morphological case) in the output,
while ‘IDENT [Case]’ requires the case feature values in the input not to be
different from the case value in the output. It is important to keep in mind
that ‘IDENT [Case]’ amounts to a local conjunction of ‘IDENT [Gen, Acc, Dat]’
and ‘IDENT [Gen, Acc, Dat],’ which means that ‘IDENT [Case]’ is satisfied
when a distinct accusative or ergative case morpheme is available in the
presence of a distinct genitive case morpheme.47 This is derived from (41a),
which requires a distinct accusative and/or ergative case morpheme in the
presence of a distinct genitive case morpheme.48
The third step is to derive the patterns of case syncretism in (42) from the
interaction of the markedness and faithfulness constraints in (57). Let us
begin with the massive syncretism in Kabardian, a Northwest Caucasian
language that involves an ergative case alignment. (58) is the constraint
ranking for Kabardian.
(58) *Gen >> {*Acc, *Erg} >> MAX [Case] >> IDENT [Case] >> *Dat
(58) ensures that the genitive and ergative case in the input are morpho-
logically realized by the dative case. This explains why actor arguments of
transitive verbs, adnominal possessors, and a wide range of oblique argu-
ments and adjuncts bear the same morphological case marking. Tableau
7.4 shows that the constraint ranking in (58) outputs the dative case
morpheme in Kabardian when it receives the syntactic ergative or genitive
case as input.

Case Assignment 353
Tableau 7.4 Case syncretism in Kabardian
ˈˈ MAX IDENT
ˈˈ
*Gen *Acc
ˈˈ *Erg [Case] [Case] *Dat
ˈˈ
ˈˈ
GEN *!
ERG ˈˈ
ˈˈ
*!
☞DAT ˈˈ
ˈˈ
* *
NOM
ˈˈ *!
ˈ
Tableau 7.5 Case syncretism in Old Persian
MAX IDENT ˈˈ
ˈˈ
*Gen [Case] [Case] *Acc
ˈˈ *Erg *Dat
ˈˈ
ˈˈ
GEN *!
ACC ˈˈ
ˈˈ
* *!
☞DAT ˈˈ
ˈˈ
* *
NOM *! ˈˈ
ˈ
Re-ranking of ‘MAX [Case]’ and ‘IDENT [Case]’ as shown in (59) yields the
genitive-dative syncretism, illustrated by Old Persian.
(59) *Gen >> MAX [Case] >> IDENT [Case] >> {*Acc, *Erg} >> *Dat
The constraint ranking in (59) yields an accusative or ergative case system in

which the dative and genitive cases receive the same morphological
marking. Tableau 7.5 shows that the genitive case in the input is mapped
to the dative case morpheme in Old Persian.
The hierarchized markedness constraints in (57a) require genitive to turn
into dative when ‘*Gen’ outranks the two faithful constraints, since dative is
the least-marked non-nominative case morpheme. This raises the question of
how to handle the genitive-ergative and genitive-accusative syncretism, since
no matter how the constraints in (57a)–(57c) may be ranked, their ranking
cannot output ergative or accusative when it receives genitive as input.
For the purpose of illustration, let us examine the Tagalog case-marking
system, which displays both the accusative-dative and genitive-ergative syn-
cretism. The first point to note is that the accusative-dative syncretism is
derived from the following constraint ranking.
(60) MAX [Case] >> IDENT [Case] >> *Gen >> *Acc >> *Erg >> *Dat
Given that ‘IDENT [Case]’ requires a distinct accusative and/or ergative case
morpheme in the presence of a distinct genitive case morpheme and that
faithful realization of accusative incurs a more serious violation than that of
ergative in (60), (60) leads accusative to be syncretized with dative, while
requiring ergative to be realized faithfully, as shown in Table 7.6.49
The next question to ask is how to derive the genitive-ergative syncretism.
It is important to note that no matter how the constraints in (57) may be

354 WATARU NAKAMURA
Table 7.6 Syntax–morphology mapping in Tagalog
Syntax GEN ERG ACC DAT NOM
Morphology Gen Erg Dat Nom

----------------------------------------------------------------------------------
Case morpheme ng/ni sa/kay ang/si
ranked, it is impossible to turn the genitive into the ergative. The reason is,
simply, that the dative case always beats the ergative case. This leaves us
with no choice but to postulate a case morpheme that underspecifies its case
value, so that it may be compatible with the genitive or ergative.50 The above
discussion suggests that ng/ni, sa/kay and ang/si constitute a case morpheme
with no inherent case value, a dative case morpheme and a nominative case
morpheme, respectively.
The major piece of evidence for analysing ng/ni as having no inherent case
value comes from the fact that they mark not only A, O and S arguments in
voiceless constructions but also some instrumental nouns, as illustrated by
(61a,b) (Kroeger 1993: 45).
(61) Tagalog (Kroeger 1993: 45)

a. Binalutan niya ng¼papel ang¼libro.
pfv.wrap.lv 3sg.gen gen¼paper nom¼book
‘He covered the book with paper.’
b. Dadalhin ko ng¼sipit ang¼isda
fut.bring.pv 1sg.gen gen¼chopsticks nom¼fish
sa¼mesa.
dat¼table
‘I’ll take the fish to the table with chopsticks.’
The fact that ng/ni-marked nouns include, but are not restricted to, actor
arguments of transitive verbs and adnominal possessors demonstrates that
ng/ni lack any inherent case value and that ng/ni functions as the default
(elsewhere) case morpheme.51
An analogous account can be given of the genitive-accusative syncre-
tism in Northern Sámi, which allows its genitive/accusative case mor-
pheme to mark complement nouns of both postpositions and
prepositions, subject arguments of non-finite clauses, and nouns that
represent a standard of comparison in comparative constructions (see
Section 7.5.3) in addition to undergoer arguments of transitive verbs
and adnominal possessors.
(62) Northern Sámi (Valijärvi and Kahn 2017: 51, 52)

a. Oahppit lohket ollu girjjiid.
student.nom.pl read.prs.3pl a.lot book.gen.pl
‘Students read a lot of books.’
b. Beatnaga namma lea Čáhppe.
dog.gen.sg name.nom.pl is Čáhppe
‘The dog’s name is Čáhppe.’

Case Assignment 355
c. Máret orru vánhemiid luhtte.

Máret lives parents.gen.pl with
‘Máret lives with her parents.’ (¼(47a))
d. Son ođii Máhte humadettiin.
3sg.nom slept Máhtte.gen.sg talk.ger
‘S/he slept while Máhtte was talking.’ (¼(47b))
e. Biila lea skohtera ođđaseabbo.
car.nom.sg is snowmobile.gen.sg newer
‘The car is newer than the snowmobile.’ (¼(47c))
(62a,b) and (47a)–(47c) (repeated above as (62c)–(62e)) show that the genitive/
accusative case morpheme in Northern Sámi behaves rather as the default
case morpheme and hence has no inherent case value of its own.52
The genitive-accusative syncretism also obtains in Estonian. The genitive/
accusative case morpheme in Estonian not only marks adnominal posses-
sors and some undergoer arguments of transitive verbs (see Section 7.5.2),
but also modifiers of adjectives, complement nouns of most postpositions
and some prepositions, and measure adverbials that are singular (Ehala
1994: 180–181; Miljan 2008: 176). The latter two uses are illustrated in
(63a)–(63d).
(63) Estonian (Miljan and Cann 2013: 351, 363)

a. poisi kõrvale
boy.gen.sg beside.all
‘(to) beside the boy’
b. Ta viibis Londonis nädala/nädalaid.
3sg.nom stay.pst.3sg London.ine week.gen.sg/prt.pl
‘S/he stayed in London for a week/for weeks.’
c. Raul ehitas suvila.
Raul.nom build.pst.3sg cottage.gen.sg
‘Raul built a cottage.’ (¼(48b)) (completed)
d. Raul ehitas suvila-t.
Raul.nom build.pst.3sg cottage-prt.sg
‘Raul was building a cottage.’ (¼(48c)) (incomplete)
The case alternation in (63b) shows that measure adverbials that are plural
receive partitive case, while those that are singular receive genitive case.53
I follow Miljan and Cann (2013) in defining the genitive/accusative case in
Estonian as involving ‘non-subject’ dependency on some head (i.e. a verb,
noun or adposition) and deriving its contextual interpretation as found in
(63b) and (48b) (repeated as (63c), which is in contrast with (63d)) (quantitative
specification and aspectual boundedness) inferentially from its paradigmatic
contrast with the partitive case (which Miljan and Cann (2013) analyse as
carrying the meaning of ‘part of’; cf. Brattico 2011; Metslang 2014; Norris
2018). Their analysis of the genitive/accusative case morpheme in Estonian as
being ‘structural’ amounts to defining it as having no case value.
To summarize, this subsection has shown that there are two types of case
syncretism, one that involves markedness reduction and is derived from

356 WATARU NAKAMURA
ranking the constraints in (57) (i.e. the genitive-dative, genitive-ergative/

accusative-dative, and accusative-dative syncretism) and the other that
cannot be derived from any ranking of the constraints in (57) (i.e. the
genitive-ergative and the genitive-accusative syncretism) and necessitates
introduction of the default case morpheme with no case value of its own.
It has used the Tagalog, Northern Sámi and Estonian examples to illustrate
how the latter type of syncretism is derived by introducing the default case
morphemes that are assigned to any NP that would be caseless otherwise.
7.6 Conclusion
This chapter has outlined the RRG theory of case assignment in its original
formulation (Van Valin 1991, 1993, 2005, 2009; Van Valin and LaPolla 1997),
its OT implementation (Nakamura 1999a, 1999b) and its extension to instru-
mental case assignment and case syncretism (Nakamura 2011, 2015, 2021).
The original RRG case theory is a version of dependent case theory that is
dependent not on phrase-structural asymmetry but on the ranking of actor
and undergoer, while its OT implementation not only defines nominative
and absolutive (conflated as the any-argument case despite their distribu-
tional difference), accusative and ergative case with no appeal to the
ranking of macroroles, but also derives the typological variation of case
systems from re-ranking of the four simple constraints.
Assuming either version of the RRG case theory, the chapter has also
outlined two attempts to further extend its scope: (i) providing the new
macrorole-based definition of instrumental case and deriving the diverse
interpretations of instrumental-marked nouns from the interaction of its
monosemous meaning with various contextual information rather than
from the alleged polysemy of instrumental case (Wierzbicka 1980; Janda
1993; Narrog and Ito 2007) and (ii) deriving the typological variation of case
syncretism among dative, accusative, ergative and genitive from the compe-
tition of the markedness constraints derived from the Case Hierarchy (Sil-
verstein 1976, 1980/1993) and the faithfulness constraints and the default
case morphemes with no case value (Nakamura 2002, 2011).
References
Agnihotri, Rama Kant. 2007. Hindi: An Essential Grammar. London: Routledge.

Aikio, Ante. 2009. The Structure of North Saami. Course handout, Department
of Linguistics, University of Utah.
Aldai, Gontzal. 2008. From ergative case marking to semantic case marking:
The case of historical Basque. In Mark Donohue and Søren Wichmann
(eds.), The Typology of Semantic Alignment, 197–218. Oxford: Oxford Univer-
sity Press.

Case Assignment 357
Aldai, Gontzal. 2009. Is Basque morphologically ergative? Western Basque vs.

Eastern Basque. Studies in Language 33: 783–831.
Andrews, Avery D. 1982. The representation of case in Modern Icelandic. In
Joan Bresnan (ed.), The Mental Representation of Grammatical Relations,
427–503. Cambridge, MA: MIT Press.
Andrews, Avery D. 1990. Case structure and control in Modern Icelandic. In
Joan Maling and Annie Zaenen (eds.), Syntax and Semantics 24: Modern
Icelandic Syntax, 187–234. New York: Academic Press.
Baker, Mark. 2015. Case: Its Principles and Its Parameters. Cambridge: Cam-
Benvenuto, Maria Carmela and Flavia Pompeo. 2015. The Old Persian geni-
tive: A study of a syncretic case. In Anna Krasnowolska and Renata Rusek-
Kowalska (eds.), Studies on the Iranian World I: Before Islam, 13–29. Krakow:
Jagiellonian University Press.
Berlin, Brent and Paul Kay. 1969. Basic Color Terms: Their Universality and
Evolution. Berkeley, CA: University of California Press.
Blake, Barry J. 2001. Case. Cambridge: Cambridge University Press.
Blansitt, Edward L., Jr. 1988. Datives and allatives. In Michael Hammond,
Edith A. Moravcsik and Jessica Wirth (eds.), Studies in Syntactic Typology,
Brattico, Pauli. 2011. Case assignment, case concord, and the quantifica-
tional case construction. Lingua 121: 1042–1066.
Bricker, Victoria. 1981. The source of the ergative split in Yucatec Maya.
Journal of Mayan Linguistics 2: 83–127.
Caha, Pavel. 2013. Explaining the structure of case paradigms by the mech-
anisms of Nanosyntax: The Classical Armenian nominal declension. Nat-
ural Language and Linguistic Theory 31: 1015–1066.
Calabrese, Andreas. 1995. A constraint-based theory of phonological mark-
edness and simplification procedures. Linguistic Inquiry 26: 373–463.
Colarusso, John. 1992. A Grammar of the Kabardian Language. Calgary: Univer-
sity of Calgary Press.
Comrie, Bernard. 1981. The Languages of the Soviet Union. Cambridge: Cam-
Coon, Jessica. 2009. Comments on Austronesian nominalism: A Mayan per-
spective. Theoretical Linguistics 35: 73–93.
Coon, Jessica. 2013. Aspects of Split Ergativity. Oxford: Oxford University
Press.
Cornilescu, Alexandra. 2000. Notes on the interpretation of the prepos-
itional accusative in Romanian. Bucharest Working Papers in Linguistics 2:
91–106.
Croft, William. 1991. Syntactic Categories and Grammatical Relations: The Cogni-
tive Organization of Information. Chicago: University of Chicago Press.
Dowty, David. 1991. Thematic proto-roles and argument selection. Language
67: 547–619.

358 WATARU NAKAMURA
Durie, Mark. 1985. A Grammar of Acehnese on the Basis of a Dialect of North Aceh.
Dordrecht: Foris
Durie, Mark. 1987. Grammatical relations in Acehnese. Studies in Language 11:
365–399.
Ehala, Martin. 1994. Russian influence and the change in progress in the
Estonian adpositional system. Linguistica Uralica 3: 177–193.
Farrell, Patrick. 2009. The preposition with in Role and Reference Grammar.
In Lilián Guerrero, Sergio Ibáñez and Valeria A. Belloro( eds.), Studies in Role
and Reference Grammar, 179–202. Mexico City: UNAM Press.
Fowler, George. 1996. Oblique passivization in Russian. The Slavic and East
European Journal 40: 519–545.
Georgopoulos, Carol. 1991. Syntactic Variables: Resumptive Pronouns and A’
Binding in Palauan. Dordrecht: Kluwer Academic Publishers.
Gerdts, Donna B. 1988. Object and Absolutive in Halkomelem Salish. New York:
Garland.
Gönczöl-Davies, Ramona. 2008. Romanian: An Essential Grammar. London:
Routledge.
Hale, Ken. 1983. Warlpiri and the grammar of non-configurational lan-
guages. Natural Language and Linguistic Theory 1: 5–47.
Hill, Virginia. 2013. The direct object marker in Romanian: A historical
perspective. Australian Journal of Linguistics 33: 140–151.
de Hoop, Helen and Andrej L. Malchukov. 2008. Case-marking strategies.
Linguistic Inquiry 39: 565–587.
Huumo, Tuomas. 1996. Domain shifts and the grammaticalization of case:
A case study of the Finnish adessive. Folia Linguistica Historica 30: 73–96.
Inaba, Nobufumi and Rogier Blokland. 2001. Allative, genitive, and partitive:
On the dative in Old Finnish. Congressus Nonus Internationalis Fenno-Ugris-
tarum 7, Pars IV: 421–431.
Jakobson, Roman. 1936/1984. Beitrage zur allgemeinen Kasuslehre:
Gesamtbedeutungen der russischen Kasus. Travaux du Cercle Linguistique
de Prague 6: 240–288. Reprinted in Linda R. Waugh and Morris Halle
(eds.), 1984. Russian and Slavic Grammar: Studies 1931–1981, 59–103. Berlin:
Mouton.
Janda, Laura A. 1993. The Geography of Case Semantics: The Czech Dative and the
Russian Instrumental. Berlin: Mouton de Gruyter.
Johns, Alana. 1992. Deriving ergativity. Linguistic Inquiry 23: 57–87.
Jolly, Julia A. 1991. Prepositional Analysis within the Framework of Role and
Reference Grammar. New York: Peter Lang.
Jolly, Julia A. 1993. Preposition assignment in English. In Robert D. Van
Valin, Jr. (ed.), Advances in Role and Reference Grammar, 275–310. Amsterdam:
John Benjamins.
Josephs, Lewis S. 1975. Palauan Reference Grammar. Honolulu: University Press
of Hawaii.

Case Assignment 359
Karlsson, Fred. 1999. Finnish: An Essential Grammar. London: Routledge.

Kaufman, Daniel. 2009. Austronesian nominalism and its consequences:
A Tagalog case study. Theoretical Linguistics 35: 1–49.
Kay, Paul and Luisa Maffi. 1999. Color appearance and the emergence and
evolution of basic color lexicons. American Anthropologist 101: 743–760.
Kilby, David A. 1977. Deep and Superficial Cases in Russian. Munich: Kubon and
Sagner.
Kiparsky, Paul. 2001. Structural case in Finnish. Lingua 111: 315–376.
Koenig, Jean-Pierre, Gail Mauner, Breton Bienvenue and Kathy Conklin. 2008.
What with? The anatomy of a (proto)-role. Journal of Semantics 25: 175–220.
Koivisto, Vesa. 2017. The essive in Karelian. In Casper de Groot (ed.), Uralic
Essive and the Expression of Impermanent State, 161–184. Amsterdam: John
Benjamins.
Korhonen, Mikko. 1991[1996]. Remarks on the structure and history of the
Uralic case system. Suomalais-Ugrilaisen Seuran Aikakauskirja 83: 163–180.
Reprinted in Tapani Salminen (ed.), 1996. Typological and Historical Studies in
Language by Mikko Korhonen, 219–242. Helsinki: Suomalais-Ugrilainen
Seura.
Kroeger, Paul. 1993. Phrase Structure and Grammatical Relations in Tagalog.
Stanford, CA: CSLI Publications.
Latrouite, Anja. 2011. Voice and Case in Tagalog: The Coding of Prominence and
Orientation. Unpublished PhD dissertation, Heinrich Heine University
Düsseldorf.
Levin, Beth and Malka Rappaport Hovav. 2005. Argument Realization. Cam-
Lindström, Liina and Virve-Anneli Vihman. 2017. Who needs it? Variation in
experiencer marking in Estonian ‘need’-constructions. Journal of Linguistics
53: 1–34.
McCarthy, John J. 2008. Doing Optimality Theory: Applying Theory to Data.
Oxford: Blackwell.
Maling, Joan. 1993. Of nominative and accusative: The hierarchical assign-
ment of grammatical case in Finnish. In Anders Holmberg and Urpo
Nikanne (eds.), Case and Other Functional Categories in Finnish Syntax, 49–74.
Berlin: Mouton de Gruyter.
Marantz, Alec. 1991. Case and licensing. Proceedings of the Eastern States Confer-
ence on Linguistics 8: 234–253.
Mardale, Alexandru. 2008. Microvariation within differential object
marking: Data from Romance. Revue Roumaine de Linguistique 53: 449–467.
Matsumura, Kazuto. 1994. Is the Estonian adessive really a local case? Journal
of Asian and African Studies 46/47: 223–235.
Matsumura, Kazuto. 1996. The dative use of the adessive case in Estonian:
A corpus-based study. In Kazuto Matsumura and Tooru Hayashi (eds.), The
Dative and Related Phenomena, 31–79. Tokyo: Hituzi Shobo.
Mazzitelli, Lidia Federica. 2017. Predicative possession in the languages of
the Circum-Baltic area. Folia Linguistica 51: 1–60.

360 WATARU NAKAMURA
Metslang, Helena. 2014. Partitive noun phrases in the Estonian core argu-
ment system. In Silvia Luraghi and Tuomas Huumo (eds.), Partitive Cases
and Related Categories, 177–256. Berlin: Mouton de Gruyter.
Metslang, Helle and Matti Erelt. 2006. Estonian clause patterns: From Finno-
Ugric to Standard Average European. Linguistica Uralica 42: 254–266.
Miljan, Merilin. 2008. Grammatical Case in Estonian. Unpublished PhD disser-
tation, University of Edinburgh.
Miljan, Merilin and Ronnie Cann. 2013. Rethinking case marking and case
alternation in Estonian. Nordic Journal of Linguistics 36: 333–379.
Mohanan, Tara. 1994. Argument Structure in Hindi. Stanford, CA: CSLI
Publications.
Nakamura, Wataru. 1999a. Functional Optimality Theory: Evidence from
split case systems. In Michael Darnell et al. (eds.), Functionalism and Formal-
ism in Linguistics, Vol. 2: Case Studies, 253–276. Amsterdam: John
Benjamins.
Nakamura, Wataru. 1999b. An Optimality-Theoretic account of the Japanese
case system. Studies in Language 23: 607–660.
Nakamura, Wataru. 2002. Deriving morphological cases: Markedness consid-
erations. In Reinhard Rapp (ed.), Sprachwissenschaft auf dem Weg in das dritte
Jahrtausend. Akten des 34. Linguistischen Kolloquiums in Germersheim 1999, Teil
1, 151–158. Frankfurt am Main: Peter Lang.
Nakamura, Wataru. 2008. Fluid transitivity and generalized semantic roles.
In Robert D. Van Valin, Jr. (ed.), Investigations of the Syntax–Semantics–
Pragmatics Interface, 101–116. Amsterdam: John Benjamins.
Nakamura, Wataru. 2011. Case syncretism in typological perspective: An
RRG-OT account. In Wataru Nakamura (ed.), New Perspectives in Role and
Reference Grammar, 35–63. Newcastle upon Tyne: Cambridge Scholars
Publishing.
Nakamura, Wataru. 2015. A neo-Jakobsonian account of default oblique
cases: Instrumental vs. dative. Paper read at the 2015 International Con-
ference on Role and Reference Grammar, Heinrich Heine University
Düsseldorf.
Nakamura, Wataru. 2016. A two-tiered theory of case features: The case of
the Hindi case(-marking) system. Paper presented in the 49th Annual Meet-
ing of the Societas Linguistica Europaea, University of Naples Federico II.
Nakamura, Wataru. 2021. A neo-Jakobsonian account of default oblique
cases: Instrumental vs. dative. In Robert D. Van Valin, Jr. (ed.), Challenges
in the Analysis of the Syntax–Semantics–Pragmatics Interface. Newcastle upon
Narasimhan, Bhuvana. 1998. A lexical-semantic explanation for ‘quirky’ case
marking in Hindi. Studia Linguistica 52: 48–76.
Narrog, Heiko and Shinya Ito. 2007. Re-constructing semantic maps: The
comitative-instrumental area. Sprachtypologie and Universalienforschung 60:
273–292.

Case Assignment 361
Nelson, Diane. 1998. Grammatical Case Assignment in Finnish. New York:

Garland.
Nelson, Diane. 2003. Case and adverbials in Inari Saami and Finnish. Nordlyd
31: 708–722.
Nevins, Andrew. 2007. The representation of third person and its conse-
quences for person-case effects. Natural Language and Linguistic Theory 25:
273–313.
Norris, Mark. 2018. Unmarked case in Estonian nominals. Natural Language
and Linguistic Theory 36: 523–562.
Oniga, Renato. 2014. Latin: A Linguistic Introduction. Oxford: Oxford University
Press.
Oranen, Nicklas. 2019. On the use of the adessive case in Tver Karelian.
Lähivõrdlusi 29: 204–227.
Preminger, Omer. 2012. The absence of an implicit object in unergatives:
New and old evidence from Basque. Lingua 122: 278–288.
Prince, Alan and Paul Smolensky. 2004. Optimality Theory: Constraint Interaction
in Generative Grammar. Oxford: Basil Blackwell.
Rezac, Milan, Pablo Albizu and Ricardo Etxepare. 2014. The structural erga-
tive of Basque and the theory of case. Natural Language and Linguistic Theory
32: 1273–1330.
Rissman, Lilia. 2010. Instrumental with, locatum with, and the argument/
adjunct distinction. LSA Annual Meeting Extended Abstracts 2010.
Sabbagh, Joseph. 2016 Specificity and objecthood in Tagalog. Journal of Lin-
guistics 52: 639–688.
Sadakane, Kumi and Masatoshi Koizumi. 1995. On the nature of the ‘dative’
particle ni in Japanese. Linguistics 33: 5–33.
Schultze-Berndt, Eva and Nikolaus P. Himmelmann. 2004. Depictive sec-
ondary predicates in cross-linguistic perspective. Linguistic Typology 8:
59–131.
Seržant, Ilja A. 2015. Dative experiencer constructions as a Circum-Baltic
isogloss. In Peter Arkadiev, Axel Holvoet and Björn Wiemer (eds.), Contem-
porary Approaches to Baltic Linguistics, 325–348. Berlin: Mouton de Gruyter.
Dixon (ed.), Grammatical Categories in Australian Languages, 112–171. Can-
berra: Australian Institute of Aboriginal Studies.
Silverstein, Michael. 1980/1993. Of nominatives and datives: Universal Gram-
mar from the bottom up. Unpublished manuscript. Reprinted in Robert D.
Van Valin, Jr. (ed.), 1993. Advances in Role and Reference Grammar, 465–498.
Smith, Henry. 1996. Restrictiveness in Case Theory. Cambridge: Cambridge
University Press.
Smolensky, Paul. 1995. On the Internal Structure of the Constraint Component of
UG. Presented at University of California, Los Angeles.

362 WATARU NAKAMURA
Starosta, Stanley, Andrew K. Pawley and Lawrence A. Reid. 1982. The evolu-
tion of focus in Austronesian. In Amram Halim, Lois Carrington and S. A.
Wurm (eds.), Papers from the Third International Conference on Austronesian
Linguistics, Vol. 2: Tracking the Travelers, 145–170.
Stolz, Thomas, Cornelia Stroh and Aina Urdze. 2006. On Comitatives and
Related Categories: A Typological Study with Special Focus on Languages of Europe.
Sulkala, Helena and Merja Karalainen. 1992. Finnish. London: Routledge.
Tamm, Anne. 2007. Perfectivity, telicity, and Estonian verbs. Nordic Journal of
Linguistics 30: 229–255.
Tomić, Olga Mišeska. 2004. The Balkan Sprachbund properties: An introduc-
tion. In Olga Mišeska Tomić (ed.), Balkan Syntax and Semantics, 1–55. Amster-
Trask, Robert L. 1979. On the origins of ergativity. In Frans Plank (ed.),
Ergativity: Toward a Theory of Grammatical Relations, 385–404. London: Aca-
demic Press.
Tsunoda, Tasaku. 1981. The Djaru Language of Kimberley, Western Australia.
Canberra: Pacific Linguistics.
Valijärvi, Riita-Liisa and Lily Kahn. 2017. North Sámi: An Essential Grammar.
London: Routledge.
Van Valin, Robert D., Jr. 1991. Another look at Icelandic case marking and
grammatical relations. Natural Language and Linguistic Theory 9: 145–194.
Robert D. Van Valin, Jr. (ed.), Advances in Role and Reference Grammar, 1–164.
Van Valin, Robert D., Jr. 2009. Case in role and Reference Grammar. In
Andrej Malchukov and Andrew Spencer (eds.), The Oxford Handbook of Case,
102–120. Oxford: Oxford University Press.
Van Valin, Robert D., Jr. 2012. Lexical representation, co-composition, and
Timberlake (eds.), Language Typology and Historical Contingency, 91–123.
Van Valin, Robert D., Jr. 2018. Some issues regarding (active) accomplish-
and Expanding Role and Reference Grammar, 71–93. Freiburg: Albert-Ludwigs-
Universität, Universitätsbibliothek.
Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning,

Case Assignment 363

Press.
Wierzbicka, Anna. 1980. The Case for Surface Case. Ann Arbor, MI: Karoma
Publishers.
Wiltschko, Martina. 2006. On ‘ergativity’ in Halkomelem Salish. In Alana
Johns, Diane Massam and Juvenal Ndayiragije (eds.), Ergativity, 197–227.
Dordrecht: Springer.
Yip, Moira, Joan Maling and Ray Jackendoff. 1987. Case in tiers. Language 63:
217–250.
Zagona, Karen. 2001. The Syntax of Spanish. Cambridge: Cambridge University
Press.
Notes
1 See Levin and Rappaport Hovav (2005: 51–68) for a critical comparison of
macroroles with proto-roles (Dowty 1991).
2 The underlying assumption is that the relative degree of affectedness
determines which non-actor argument serves as undergoer. See Van Valin
(2005: 109–115) for further discussion.
3 Van Valin (1991, 2005) attributes the idea of dative being the default case
to Silverstein (1980/1993).
4 For example, Japanese allows source arguments of ditransitive verbs such
as morau ‘receive’ and kariru ‘borrow’ to be marked by the dative (as well as
the ablative) case particle.
5 See Jolly (1991: 115–119) for an LS-based account of some other non-
locative uses of from (e.g. John suffered from arthritis).
6 German behaves like Icelandic with respect to case preservation under
passivization, while Japanese allows no such case preservation. See
Section 7.4.1 for an OT reformulation of the case assignment rules in (9)
and how it accounts for the contrast between Icelandic/German and
Japanese with respect to case preservation.
7 Predicate adjectives that modify covert controllee arguments of depend-
ent cores in Icelandic control constructions may receive nominative case
or the case of the controller argument in the matrix core (Andrews 1982:
450–456). This fact shows that the controllee argument may receive nom-
inative case and agree in case with predicate adjectives that modify it. We
may take this agreement fact as another piece of evidence that the DCA
for Icelandic is the core.
8 Matrix-coding constructions are also attested in ergative and active lan-
guages. For example, Basque requires actor arguments of transitive verbs
in non-finite gerundive complements of perception verbs to receive abso-
lutive case (Rezac et al. 2014).

364 WATARU NAKAMURA
9 The auxiliary verb in (24b) (d-u-ø) contains the absolutive affix d-, which
refers to no clausal argument. The verb dantzatu is a denominal verb that
incorporates dantza ‘dance’ and the absolutive affix refers to this incorp-
orated noun. See Preminger (2012) for related discussion.
10 Aldai’s argument is based on the case marking of S arguments of lexically
simple unergative verbs. See Aldai (2009: 797–800) for discussion of com-
pound unergative predicates formed with egin ‘do, make’ and a few other
light verbs. The fact that Basque has an ergative dialect in addition to an
active dialect justifies analysing the actor and undergoer markers in the
Western dialect as ergative and absolutive, respectively.
11 See Farrell (2009) for an exception to the ICR.
12 It is questionable whether (28) is in a subset relation with (9c). Rissman
(2010) compares the locatum with-phrases with the instrumental with-
phrases and shows that the former are syntactic arguments, while the
latter are adjuncts. See also Koenig et al. (2008) for related discussion of
the semantic and syntactic status of implement with-phrases.
13 Van Valin and LaPolla (1997: 381–382) attempt to extend the ICR to the
uses of with that mark manner adverbials in English (e.g. with enthusiasm,
with difficulty).
14 See de Hoop and Malchukov (2008) for an alternative OT-based case theory.
15 Russian allows some instrumental and genitive-marked arguments (but
not any dative-marked arguments) to receive nominative case under
passivization (Fowler 1996). One may be tempted to interpret this fact
as an indication that Russian also belongs to (33b) (under the assumption
that non-subject genitive/instrumental-marked arguments are not
undergoers), but this move would leave it unexplained why dative-
marked arguments may not undergo passivization in Russian.
16 It remains an open question whether or not there are any other non-
accusative case systems that do not allow case preservation under
passivization.
17 An asterisk mark (*) indicates a violation incurred by each candidate for a
given constraint. The asterisk followed by the exclamation mark indi-
cates a fatal violation and the candidate that incurs that violation is
eliminated from further consideration. The shaded cells indicate that
they exert no influence on the selection of the optimal output(s).
18 What is termed ‘instrumental’ in (35b) includes the ablative case in
Latin, which combines three formerly distinct cases, instrumental, loca-
tive and ablative proper (Oniga 2014: 254).
19 See Van Valin (2005: 111–112) for more relevant data from Croatian
and Dyirbal.
20 The instrumental adposition/case has no LS of its own and therefore has
no argument slot for a noun it marks (Van Valin and LaPolla 1997; Van
Valin 2005).
21 The notion of QS was incorporated into RRG by Van Valin and
LaPolla (1997).

Case Assignment 365
22 The fact that the possessive relation between Petr and his body part is
permanent and holds outside the event of nodding indicates that the
main verb kivnutʹ ‘nod’ and covert possessive predicate do not form a
complex nucleus.
23 (34c) is comparable to secondary predicate constructions, which describe
the (resulting) state of a clausal participant and involve a non-
subordinate core juncture (Schultze-Berndt and Himmelmann 2004).
They share the same juncture–nexus type, but the difference lies in the
fact that the possessum noun in (34c) specifies which part of the posses-
sor is directly involved in the event instead of representing a temporary
state of the possessor.
24 We may compare the two types of dative in (41a,b) to the two types of the
dative case particle ni in Japanese (Sadakane and Koizumi 1995): ni as a
(non-predicative) case marker and ni as a predicative postposition. See also
Blansitt (1988), which may be taken as an attempt to elaborate (41b) by
establishing implicational relations among dative, allative and locative.
25 Silverstein (1980/1993) draws an explicit analogy between the develop-
ment of case-marking systems and that of the colour-term systems
(Berlin and Kay 1969; Kay and Maffi 1999).
26 The genitive case clitic in Hindi may case-mark subjects of a small
number of intransitive verbs (Mohanan 1994) despite its adjective-like
declension (it agrees in gender, number and case with a possessum noun
it occurs with) (Agnihotri 2007: 187–189). The chimeric status of this
genitive clitic is indicated in (42i) by putting ‘Gen’ in parentheses.
27 The morpheme a in (44) is an NP marker and is not glossed (Georgopou-
los 1991: 237).
28 Personal pronouns preserve a four-way case distinction (i.e. nominative,
accusative, dative and genitive) (Gönczöl-Davies 2008). We can derive the
lack of accusative marking on some transitive undergoer arguments
from the definiteness/animacy-based split accusativity.
29 Third-person agreement markers in matrix indicative clauses alone oper-
ate on an ergative-absolutive basis (Wiltschko 2006).
30 For simplicity, in this paper, I gloss the genitive/accusative case mor-
phemes in the examples taken from Northern Sámi, Estonian and Finn-
ish as ‘genitive.’
31 See Tamm (2007) for a detailed semantic account of the alternation
between the partitive and accusative/nominative case on undergoer
arguments of transitive verbs. See also Miljan and Cann (2013: 365–372)
for how the notion of aspectual boundedness arises.
32 The genitive case marker turns out to be neutral to the distinction
between specific and non-specific, since it may mark specific common
nouns and pronouns in voiceless constructions, as illustrated by (50g,h).
33 See Latrouite (2011: Ch. 5) and Sabbagh (2016: 656–660) for further
discussion of the clause-internal distribution of ng/ni and the case alter-
nation between ng/ni and sa/kay.

366 WATARU NAKAMURA
34 Kroeger (1993: 50) notes that gerundive constructions also bear no voice
marking on the verb stem and allow only what he terms ‘default
case assignment’.
35 Recent-perfective constructions are a type of impersonal (or ‘pivotless’)
construction. See Kroeger (1993: 48–51) for a list of impersonal construc-
tions in Tagalog.
36 An ergative-dative syncretism is hardly attested despite the fact that it is
predicted to be possible by the CH. The cross-linguistic paucity of the
ergative-dative syncretism requires a principled explanation. Nakamura
(2011) attributes its near unavailability to an independent semantic con-
straint to the effect that the farther one of the co-arguments of a verb is
away from the other one on the causal chain, the less likely they are to
receive the same morphological marking. The essential idea is that the CH
works in tandem with this semantic constraint (adapted from Croft (1991:
Ch. 5)) to constrain the possible patterns of case syncretism.
37 Aikio (2009) notes that Máhtes in (53g) does not serve as an
adnominal possessor.
38 Some verbs subcategorize for a locative-marked argument that resists
any immediate spatial interpretation (e.g. heaitit ‘stop’, dolkat ‘get sick of’,
beroštit ‘care about’). The same holds for verbs with an illative-marked
argument (Valijärvi and Kahn 2017: 54, 56).
39 (53h) is a crucial example, since it shows that the locative case can be
treated on a par with the nominative case for the purpose of indicating
the relative degree of agentivity.
40 The external local cases (allative, adessive and ablative) have a much
stronger tendency to mark non-spatial relations than the corresponding
internal local cases (illative, inessive and elative) in Estonian. As a case in
point, Matsumura (1994) compares the adessive case with the inessive
case in terms of the frequencies of three semantic types, (i) place, (ii)
time, and (iii) possessor, in two text corpora and finds that the third type
accounts for more than 40% of all the adessive nouns, while the first type
that refers to a physical space accounts for less than 8%. In contrast, the
first type accounts for almost 80% of all the inessive nouns, while no
inessive noun of the third type (possessor) is found.
41 See Matsumura (1996: 40–51) and Lindström and Vihman (2017: 9–11) for
more relevant examples and further discussion of the relation between
the adessive and allative in Estonian.
42 The allative, genitive and partitive cases mark recipient arguments of
ditransitive verbs in Old Finnish (Inaba and Blockland 2001).
43 The fact that the genitive marking of sinu-n/Mati-n in (56h) is retained
under passivization and negation (Kiparsky 2001: 356–357) indicates that
they are non-macrorole core arguments.
44 Korhonen (1991/1996) notes that the adnominal uses of the genitive case
were historically derived from the Proto-Uralic -n lative case by way of its
dative (‘dative-genitive’) uses. See also Huumo (1996: 74–75, 83–86) for

Case Assignment 367
the reason the functions borne by the -n lative case have been taken over
by the external local cases.
45 Oranen (2019: 211–219) lists the functions borne by the adessive case
in Tolmači Karelian and compares its adessive case with the Finnish
and Estonian counterparts. See also Koivisto (2017: 162–164) for other
syncretic patterns (the adessive-ablative syncretism and allative-adessive-
ablative syncretism) displayed by a few other dialects of Karelian.
46 (57a)–(57c) make no reference to the nominative case value, but it is not
representationally underspecified. I follow the spirit of Calabrese (1995)
in parameterizing rules/constraints with respect to whether they may
refer to (i) all, (ii) contrastive, or (iii) marked feature values and assume
that all constraints related to case assignment may refer to contrastive
case values. It follows, then, that the nominative case value remains
invisible to constraints when it is not contrastive in the case-marking
system of a particular language (cf. Nevins 2007: 285–287).
47 Unlike ‘IDENT [Case]’, ‘MAX [Case]’ involves no local conjunction (as
shown in (57c)), since it is not sensitive to the distinctness/identity of
case morphemes and only ensures that a syntactic non-nominative case
is realized by any non-nominative case morpheme.
48 See McCarthy (2008: 214–219) for a formulation and discussion of local
conjunction (due originally to Smolensky (1995)).
49 The distinction between ng and ni is that the former marks common
nouns, while the latter marks personal names. The same contrast obtains
in sa/kay and ang/si.
50 The case-marking system in Hindi is derived from the same constraint
ranking as (60) (Nakamura 2016). The difference between Tagalog and
Hindi comes down to the availability of a case morpheme that neutral-
izes the ergative-genitive distinction.
51 This formulation accords well with Foley and Van Valin’s (1984: 389) defin-
ition of ng/ni as the default case marker for non-pivot core arguments.
52 Like Northern Sámi, Estonian uses both postpositions and prepositions,
but the latter are a relatively new development and are limited in
number. Specifically, the vast majority of postpositions in Estonian
assign genitive case to their complement nouns. In contrast, the case
assignment of prepositions is not dominated by any single case, but the
genitive case outnumbers the other cases (Ehala 1994: 180–181).
53 Measure adverbials as illustrated in (63b) exhibit the same case-marking
pattern as undergoer arguments of transitive verbs. See Maling (1993),
Nelson (1998, 2003) and Kiparsky (2001) for further illustration and
discussion of case-marked adverbials (like the one in (63b)) in the Finnic
and Sámi languages. See Van Valin (2018) for how to represent analogous
measure adverbials in terms of LS.

8
Morphology in RRG
The Layered Structure of the Word, Inflection and Derivation
Francisco J. Cortés-Rodríguez
8.1 Introduction
This chapter offers an overview of the place of inflectional and deriv-

ational morphology in Role and Reference Grammar (RRG). A number of
researchers have devoted their efforts to outlining the key topics of a
functional theory of morphology within this model. Martín Arista (2008:
126) summarizes the different approaches that would give support to
such a theory. He suggests that this would draw on structural-functional
models such as those of Dik (1997), Van Valin and LaPolla (1997) and Van
Valin (2005). Specific relevant elements in the determination of the theory
are the layering proposals for grammatical structures such as the clause
(Foley and Van Valin 1984; Hengeveld 1989) or the phrase (Rijkhoff 2002)
and the manner in which they can be extended to the word domain. Word
Syntax approaches – stemming from Marchand (1969) and continuing in
syntacticist theories of morphology such as Selkirk (1982) and Baker
(2003) – are also influential. To these, it is necessary to add the studies
that view morphology from a lexical-semantic perspective, as do Martín
Mingorance (1998 [1985]), Lieber (2004) and Štekauer (2005a, 2005b), as
they are crucial for the approach to word formation put forward in this
chapter. Nevertheless, despite its centrality within a theory of grammar,
there are still some issues to address for a fully developed morphological
theory within RRG.
In order to provide a description of the state of morphological research
within RRG, we will follow a widely accepted distinction between two
approaches in morphological studies, analytic and synthetic, as described
by Aronoff and Fudeman (2011: 12). The former deals with the description of
the internal constituency of words: analytical descriptions must state which
units are considered to be the components of morphologically complex
structures; units such as morphemes, lexemes, affixes, etc. will come into
play here.

Morphology in RRG 369
The synthetic approach has to do with the mechanisms that are invoked
by morphological models to account for the production of morphologically
complex words. From a constructivist perspective, analytical morphology
takes the perspective of the decoder, whereas when a speaker produces a
new complex word they make use of the synthetic machinery.
From the point of view of theory building and description, the analytical
approach must come first. Only after a morphological theory has established
what the building blocks of words are will it be in a position to give an
account of the manner in which such pieces are put to work to create
complex lexical units. In line with this, our description of morphological
theory within RRG will also start with an overview of the proposals offered
within the model to provide a sound description of the internal constituents
of words. This is the main goal of Section 8.2 in this chapter. Based on the
programmatic contribution by Everett (2002) and the developments into a
fully articulated model in Martín Arista (2008, 2009, 2011), we will spell out
the details of the layered structure of the word (henceforth LSW). Some
alternative views on specific aspects of Martín Arista’s proposal (such as
those expressed in Cortés-Rodríguez and Sosa 2012; Nolan 2011; Diedrichsen
2011) will also be considered.
The synthetic aspects of morphological processes are considered in
Sections 8.3 and 8.4. From a synthetic viewpoint, attention is paid to issues
such as whether there are specific morphological rules to generate complex
words or, on the contrary, they are the result of other types of processes, as
are syntactic or phonological operations. Consequently, in synthetic descrip-
tions, one must take into account whether morphology is a separate com-
ponent of grammar or just a set of phenomena assimilated to other
modules. The functional orientation of RRG has led researchers interested
in this area to place the emphasis on the external motivation of morpho-
logical phenomena. In this regard, there is widespread consensus among
RRG morphologists that lexis, syntax, semantics and pragmatics motivate
morphology.1
It is at this point, however, that inflectional and derivational morphology
take different paths, inflectional morphology aligning itself with syntactic
processes while word-formation processes are treated as essentially lexical-
semantic phenomena. Following Stump’s (2001) four-way classification of
morphological theories, inflectional morphology in RRG follows an
inferential-realizational approach. In opposition to lexical theories of inflec-
tion, in which inflectional affixes have lexical entries specifying their mor-
phosyntactic properties, an inferential theory associates morphosyntactic
properties with morphological rules, which relate a given inflected word-
form to its root. Furthermore, instead of being incremental, the RRG
approach to inflection is realizational: the association of a given root or
lexeme (e.g. work) to a set of properties like ‘3rd person singular’, ‘present
tense’ and ‘indicative mood’ licenses the attachment of the affix -s (cf. Stump
2001: 1–3).2 Derivational morphology lies outside this typology, as it is, in

370 FRANCISCO J. CORTÉS-RODRÍGUEZ
essence, a lexical phenomenon involving lexical units and the properties

encoded in their corresponding lexical entries.
In accordance with the distinction between inflectional and derivational
morphology, Section 8.3 will be devoted to the manner in which inflectional
phenomena are dealt with within the overall apparatus of RRG, while in
Section 8.4 we will turn to word-formation processes.
After the description of how inflection is encoded in the LSW, Section 8.3
offers an explanatory account of the factors that motivate inflectional
marking. The functional orientation of RRG will play a significant role in
this account. A description based on this functional standpoint presupposes
a view of morphology distributed throughout the different components of
the grammatical model. Additionally, the typological commitment of RRG
requires paying close attention to the role of inflectional processes not only
in dependent-marking languages, but also in head-marking languages, since
the interface between inflectional morphology and syntax is much tighter in
head-marking than in dependent-marking structures.
Section 8.4 will shift the focus to the domain of morphology as a lexical
phenomenon. Proposals explaining word formation as a lexicological pro-
cess involving the interaction of lexical semantics and morphology within
RRG will be discussed. The approach to derivational morphology within RRG
can be said to be markedly motivated by semantics. Word formation is
tightly connected to lexical semantics, and lexical-semantic representations,
which are at the core of RRG, become crucial for the development of a
theory of derivation. The inclusion of Qualia Theory in semantic representa-
tions provides RRG with a very robust system to account for word-formation
phenomena, in contrast to other approaches. Some conclusions are drawn in
the final section.
8.2 Foundations of RRG Morphology: The Layered

Structure of the Word
This section will establish the set of common assumptions that support the
different contributions for the development of morphology within RRG
from an analytical viewpoint. The description will proceed as follows.
Section 8.2.1 lists the basic underpinnings of the morphological theory
outlined in this chapter. Section 8.2.2 outlines the layered structure of the
word, the analytical tool responsible for grammatical analyses within the
word domain in RRG. Finally, Section 8.2.3 sketches the state of research on
operators in the LSW.
8.2.1 Theoretical Underpinnings

The goal of this section is to lay the foundations of our functional theory of
morphology. The notions upon which we will base the description of the
internal makeup of complex units are the following:3

(a) There are two types of lexical morphemes (or lexemes): free and
bound lexemes. Free lexemes belong to the major lexical categories,
whereas bound lexemes are derivational affixes. Both free and bound
lexemes will be semantically represented by means of their corres-
ponding lexical entry. Both are also grouped within the lexicon into
lexical classes defined by their similarity of meaning. The difference
between these types of lexeme lies only in their distributional
behaviour.4
(b) It follows from what is outlined in (a) that affixal derivation and
compounding are essentially the same type of phenomenon. Despite
this, in languages like Spanish or English, there are some differences
between these two types of process. Whereas the grammatical
makeup of affixal formations is quite predictable, since derivative
affixes usually specify both the type of base they are attached to
and the type of lexical category that outputs from its combination,
‘compounding is only barely syntactic’ (Jackendoff 2009: 14), since
the elements of a compound are rather opaque, in that they do not
provide this kind of specification. This is especially clear in Spanish
compounds, which can have as their nucleus either the rightmost
(as in puti-club ‘whorehouse’, lit. ‘hooker-club’) or the leftmost
constituent (as in hombre-anuncio ‘sandwich-board man’, lit. ‘man-
advertisement’).
(c) Inflectional affixes are treated differently, as they are considered to be
morphemes with no lexical status (i.e. they are not lexemes) and
therefore they are not stored in the lexicon. Apart from Nolan
(2011: 74–75), for whom inflectional affixes are allocated to a ‘mor-
pheme store’, other researchers working in RRG morphology (Everett
2002; Martín Arista 2008; Boutin 2011; Van Valin 2013) argue for an
inferential realizational approach to inflection (cf. Spencer 1998,
2004; Stump 2001): phonological operations/rules will be responsible
for providing the morphosyntactic representation of a given inflected
form.5
(d) Whereas derivation and compounding are unified, inflection and word
formation are regarded as disparate phenomena from a functional point
of view, since they serve different purposes within the overall structure
of our grammatical model. Despite this, they must be treated together at
some point, as they contribute to the formal structure of the word,
which is the maximal unit in morphology. The main theoretical con-
struct that handles the overall internal makeup of the word, and which
consequently accounts for the structural distribution of inflectional and
derivational morphology, is the layered structure of the word (LSW).
Parallel to clauses and phrases, the internal structure of words is based
on layering. The first outline of the LSW can be found in Everett (2002),
and it becomes a fully articulated model in Martín Arista’s studies (2008,
2009, 2011).

8.2.2 The Layered Structure of the Word

In a fashion similar to the layering analysis for clauses and phrases in RRG,
the LSW distinguishes three layers: the nucleus, the core and the word.
Figure 8.1 represents the LSW of a predicate of the category α as described
by Martín Arista (2009: 91).
WORD
CORE
NUCLEUS
PREDICATE
NUC OPERATORS NUCLEUS
CORE OPERATORS CORE
WORD OPERATORS WORD
Figure 8.1 The layered structure of the word (Martín Arista 2009: 91)
The LSW consists of the constituent projection for lexical constituents

(free and bound lexemes) and the operator projection for inflectional expo-
nents.6 Figure 8.2 shows the LSW of the Present-Day English genitive form
friends’, with the operators of number and case encoded in the operator
projection. A typical feature of the LSW of a simplex word is that it shows
no functions in the constituent projection. Compare this with Figure 8.3,
which analyses the Old English complex word bell-ring-estre (bell-ring-er.nom.
sg ‘bell ringer’; Martín Arista 2008: 129). In this case, the lexical constituents
bell and -estre occupy the syntactic positions of first and second argument in
the constituent projection.7 Because of this, the LSW of compound and
derived words takes the additional layer of the Complex word, which has
scope over the word layer (Martín Arista 2011: 397).
WORDN
COREN
NUCN
friends'
NUC
CORE Number
WORD Case
Figure 8.2 The LSW of friends’

COMPLEX WORDN
COREN
ARGN NUCLEUSV ARGN
WORDN WORDN
COREN COREN
NUCN NUCN
bell ring estre.NOM.SG
NUC
CORE Number
COMPLEX WORD Case

Figure 8.3 The LSW of the OE complex word bellringestre ‘bell ringer’ (Martín Arista 2008: 129)
Figure 8.4 analyses the constituent projection of the recursive formation

takal-takal mananyi (‘give out a repeated knocking’) from Pitjantjatjara-
Yankunytjatjara (Australia) (adapted from Martín Arista 2009: 110).
As is the case in the layered structure of the clause (LSC) and the layered
structure of the phrase (LSP), operators from outer layers have scope over the
inner layers. There is a further logical basis for considering inflectional
phenomena as operators in the LSW, since this is consonant with the way
in which they are analysed in the other syntactic domains in RRG, the LSC
and the LSP.
8.2.3 Operators in the LSW

Even though there is as yet no detailed description of the types of operators
that modify each of the different layers within the LSW, some general
guidelines can be offered. Drawing on data from Old and Present-Day
English, Martín Arista (2009: 93) takes the morphological status of the bases
as the central criterion for the distribution of operators: affixes that cannot
combine with derived bases encode nuclear lexical operators, whereas those
that attach to derived bases represent core lexical operators. Inflection is
also added to this as a ‘G[rammatical]-WORDop[erator]’ (Martín Arista 2008:
135). Thus, inflectional categories will at least appear distributed between

COMPLEX WORDV
COREV
ARGN NUCLEUSV
WORDN PREDICATEV
COREN
NUCN
PREDN
COMPLEX WORDN
takal-takal mananyi
Figure 8.4 The LSW of a recursive complex word (adapted from Martín Arista 2009: 110)
the core and the word lexical operators, as we see above in Figure 8.2 for
friends’. From the analysis offered in Figure 8.3 it seems clear that among the
nominal inflectional features, number (and probably gender) are core oper-
ators, whereas case has scope over the word node.8
Nuclear operators, on the other hand, will encode those features that are
not subject to percolation into higher structures (i.e. syntax-blind features).
Thus, good candidates to be classified as nuclear operators are aspectualizers
in languages where lexical aspectual features are regularly marked by
morphological means, as is typical of Slavic languages.9 Other possible
nuclear operators are some types of lexical distributive processes in Amele
(Papua New Guinea) which involve a reduplication process, as in the
following clauses (data from Roberts 2015: 19–20):10
(1) Tob-i ton-i egi-na.

ascend-pred descend-pred 3pl.sbj-prs
‘They ascend and descend.’
(2) L-i l-i h-u h-u ena.

go-pred go-pred come-pred come-pred 3pl.sbj-prs
‘He comes and goes.’
Roberts (2015: 19) explains that this type of reduplication takes place within
the verbal nucleus, in contrast with other types of reduplication processes

that take place outside the nucleus in higher layers, as shown by the
different distribution of grammatical markers.11
(3) Mel f-ec-eb f-ec-eb egi-na.

boy see-ds-3sg.su see-ds-3sg.su 3pl.sbj-prs
‘The boys are looking at each other.’
In (3) each verb has a different subject (DS) marker and 3SG subject
agreement. This is followed by verb inflection which includes 3PL agreement
with the matrix clause subject, mel ‘boys’, as well as present tense.
Table 8.1 shows a classification of the operators mentioned in the previous
description; it must be stressed that it is not an exhaustive classification and
that this is an issue that still needs further research.
Table 8.1 Operators in the LSW (a partial classification)
Layer Operator Bases Examples
Nucleus Aspectualizers (verb) Underived stems/words na-pisat (Russ. ‘write up’)

Distributives (verb) to- teran (OE ‘tear to pieces’)
Core Number (Noun) Derived book-s
Gender (Noun) vecin-a (Sp. neighbour-fem)
Word Case (Noun) Derived John’s
Diedrichsen (2011) draws attention to certain features of inflection which

must be considered. She mentions that every lexical category has its own
distinct set of inflectional categories. For instance, in German, nouns are
inflected for number, case and gender, whereas verbs inflect for tense, modal-
ity and also number and person. Furthermore, the typology of inflectional
categories and their distribution among word classes show great variability
across languages. For instance, whereas definiteness is a grammatical
category realized only on the determiner in West Germanic languages
(Diedrichsen 2011: 5), it is an inflectional category marked on the noun in
Scandinavian languages (Nübling 2008: 46, quoted also in Diedrichsen 2011:
5). Other languages are much less rigid concerning the distribution of inflec-
tional features across word classes, as illustrated by the following examples
from Bella Coola (Western Canada; data from Pavey 2010: 86–87).
(4) nuyamł-Ø ti-?immlllkı̄-tx

sing-3sg.sbj prox-boy-prox
‘the boy is singing’
(5) ?immllkı̄-Ø ti-nus?ūlχ-tx

boy-3sg.sbj prox-thief-prox
‘the thief is a boy’
(6) sx-Ø ti-nus?ūlχ-tx

bad-3sg.sbj prox-thief-prox
‘the thief is bad’

(7) ?imlk-Ø ti-nuyamł-tx

man-3sg.sbj prox-sing-prox
‘the one who is singing is a man’
(8) nus?ūlχ-Ø ‘
ti-gs-tx
thief-3sg.sbj prox-ill-prox
‘The one who is ill is a thief’
In Bella Coola, the attachment of person, number and case affixes, on the
one hand, and proximal morphemes, on the other, is flexible with regard to
class form, and is instead conditioned by the function of words in a given
sentence (Van Valin 2008). This type of phenomenon seems to run counter to
the establishment of a typologically valid arrangement of inflectional oper-
ators in the LSW (see Aronoff and Fudeman 2011: 201–202 for further
arguments on this); this is an issue which calls for further research.
8.3 The Functional Motivation of Inflectional Morphology:

Morphological and Syntactic Inflection
One of the main tenets of a functional model of language is the teleological

orientation which holds for every language phenomenon. RRG is strongly
committed to this functional prerequisite and always seeks external motiv-
ation for grammatical phenomena.12 From a synthetic perspective, this
implies that the word-internal elements described in the previous section
as components of the LSW need an external motivation, which is tanta-
mount to stating that morphological processes are visible to syntax, seman-
tics and pragmatics (Martín Arista 2011: 395).13 In essence, this section seeks
to provide an answer to the following issue posited in Van Valin (2013: 96):
How is it that syntax can target elements inside a word?
In order to provide an answer to this question, it is necessary to take into
account that the degree and modes of interaction between the LSW and
higher structures in the model is different in dependent-marking and head-
marking languages. Head-marked structures are especially challenging since
a single phonological word can at the same time be a clause, as shown in the
following example from St’át’imcets (Salishan, Canada) (Pavey 2010: 80;
from Roberts 1999: 278–279):
(9) tsún-tsi-lhkan
tell-2sg.obj-1sg.sbj
‘I told you’
From the perspective of a morphological theory in RRG, it is essential to

explain the relationship that holds between the internal structure of the
word and the internal structure of its corresponding clause.
In this section we aim to illustrate the interaction of the morphosyntactic
features encoded in the LSW with those in the layered structures of
phrases and clauses in both types of languages. As will be described, feature

percolation across the LSW and the other grammatical domains is signifi-
cantly different if we compare head- vs. dependent-marking structures.
Whereas in the latter type of language, percolation is restricted to operators
across different structures (see Section 8.3.1), in head-marking structures,
morphological features percolate up to the constituent projection in the LSC
(cf. Section 8.3.2).
8.3.1 Dependent-Marking Inflectional Morphology

Inflectional categories are grammatical markers in the structure of the word
to signal its semantic and/or pragmatic import in the construction of higher
order structures, namely phrases and clauses. Furthermore, the (complex)
word domain is parsimoniously integrated into higher grammatical
domains (clauses and phrases) by the inheritance of relational morpho-
logical features (i.e. inflection operators). As pointed out in Martín Arista
(2008: 137): ‘Inflection at Word or Complex Word Level is tantamount to
inflection at the Nucleus level of the Phrase. Consequently, the node Word
or Complex Word represents the limit of percolation of morphological
features as well as the limit of inheritance of inflectional features from
the NP.’ Figure 8.5 illustrates this.
The set of semantic markings that may be realized morphologically are
those encoded in the operator projections of both the LSC and the LSP. Thus,
the morphologically complex Spanish verb cant-aba-n (sing-ipfv.pst-3pl,
‘they were singing’) will trigger percolation of the aspect and tense features
when acting as a predicate in a clause like Los niños cantaban villancicos (‘the
children were singing carols’).
In Figure 8.6 the square brackets delimit the LSW of the word cant-aba-n
(sing-ipfv.pst-3pl, ‘they were singing’), which occupies the position of
predicate in the LSC of the sentence. There are four operators in the
LSW, encoded by two affixes, -aba, which signals both imperfective aspect
and past tense, and the suffix -n which marks both third person and plural
number. Note that not all four inflectional features are subject to percola-
tion: the -n suffix signals the person and number features that mark
agreement of the verb with the subject, but since they are not categories
that can be matched to the set of operators available in the LSC, where
the verb acts as nucleus, they are not inherited by the clause structure. On
the other hand, the imperfective and past operators are inherited by the
operator projection in the LSC of this sentence, as indicated by the dotted
lines in the figure.
8.3.2 Inflectional Morphology in Head-Marking Structures

The analyses presented in Figures 8.5 and 8.6 illustrate the behaviour of
inflectional features as encoded in the LSW in dependent-marking
languages when integrated into higher syntactic structures. In

RP
CORE
LSP
(Constituent Projection)
NUC
WN = NOUN
COREW
NUCW
LSW
W INFL percolation
NUCW
COREW
WN = NOUN
NUC
LSP CORE Number

(Operator Projection)
RP Gender
RP Case
Figure 8.5 Feature percolation in RPs (adapted from Martín Arista 2008: 136)
head-marking languages the integration of the structural domains of the

word and the clause is tighter: not only may feature percolation of the
operator projection from the LSW to the operator projection of the LSC
come into play, but some features of the LSW also occupy positions in the
constituent projection of the LSC. Van Valin (2013) offers an analysis of
morphologically complex words in head-marking languages based on Ever-
ett’s (2002) proposal for the LSW. As mentioned earlier, Everett’s model
postulates only a constituent projection in which inflectional affixes are
daughters of the lexical core, while derivation takes place within the
nucleus. As a consequence, the nucleus is opaque to syntax, but the core is
accessible to it. The morphological structure of the Lakhota (Siouan, North

SENTENCE
CLAUSE
CORE
RP NUC RP
WV = VERB
COREW
NUCW
Los niños cant aba n villancicos
NUCW
LSW
COREW imperf
WV =VERB past
WV =VERB person
WV =VERB number
NUC aspect
CORE
CLAUSE
tense
Feature Percolation LSW-LSC =

Figure 8.6 Feature percolation in the LSC
America) word Ø-wičhá-wa-k’u (inan-3pl.anim.u-1sg.a-give, ‘I gave it to

them’) would then be as shown in Figure 8.7.
The fact that in Figure 8.7 the morphological markers are part of the
constituent projection illustrates the difference between dependent- and
head-marking languages with respect to the status of inflection. In head-
marking languages, the constituent projection in the LSW will house those
inflectional features that instantiate the arguments in the semantic

COREW
ARG ARG ARG NUCW
[3SG.NMR] [3PL.ANIM.U] [1SG.A] k’u

Figure 8.7 The LSW of Lakhota verb wičhá-wa-k’u (Van Valin 2013: 113)
(NMR ¼ non-macrorole, ANIM ¼ animate, U ¼ undergoer, A ¼ actor)
SENTENCE
CLAUSE
CORE
COREV
ARG ARG ARG NUCV
[3SG.NMR] [3PL.ANIM.U] [1SG.A] k’u
PSA: NOM ACC
ACTOR UNDERGOER
[do′(1SG, Æ)] CAUSE [BECOME have′ (3PL.ANIM) (3SG)]

Figure 8.8 The LSC of Lakhota sentence wičhá-wa-k’u (Van Valin 2013: 115)
(NMR ¼ non-macrorole, ANIM ¼ animate, U ¼ undergoer, A ¼ actor, PSA ¼ privileged syntactic
argument)
representation of the core of the clause. In fact, the structure of the lexical
unit in the LSW will provide the structure of the core of the clause in the LSC;
Van Valin (2013: 115) describes that the two structures are coextensive, since
the structure of the COREw provides the structure of the core of the clause.
Note that the analysis of the COREw structure in Figures 8.7 and 8.8 is
almost identical to the structure of the core in the equivalent English clause
‘I gave it to them’. This proposal thus manages to provide comparable

structures for comparable sentences among typologically different lan-

guages while at the same time coping successfully with the differences
among such structures. In fact, the disparity between head-marking and
dependent-marking structures stems from the fact that while inflectional
features are consistently located in the operator projection of the LSC in the
case of dependent-marked constructions, they can appear distributed
between both the constituent and the operator projections in head-marked
structures. The analysis of head-marked constructions can also sit comfort-
ably within the model by Martín Arista (2008, 2009, 2011), as is shown in
Figure 8.9, which illustrates the percolation of morphological markers from
SENTENCE
CLAUSE
CORE
NUC RP RP
WV =VERB
COREW
NUCW
LSW Cant aba n villancicos
NUCW
COREW imperf
WV =VERB past
WV =VERB person: 3
WV =VERB number: PL
NUC aspect
CORE
CLAUSE tense
Figure 8.9 The LSC of Spanish sentence cantaban villancicos

the LSW to both the operator and the constituent projections in the LSC of
the Spanish clause cant-aba-n villancicos (sing-ipfv.pst-3pl carols, ‘they were
singing carols’).
Even though Spanish is mainly a dependent-marking language, it shows
subject marking on the verb by means of person and number inflectional
affixes. This makes unnecessary the presence of an independent subject
constituent, as happens in this sentence, thus allowing head-marked struc-
tures. In the absence of an overt subject argument (e.g. los niños ‘the
children’ in Figure 8.6), the suffix -n is mapped from the LSW of the verb
into the constituent projection of the LSC; this is indicated by the dotted
line in the analysis. The other morphological features in the LSW of cant-
aba-n (sing-ipfv.pst-3pl), namely imperfective and past as encoded in the
suffix -aba, percolate into the operator projection of the LSC (cf. also
Figure 8.6).
8.4 Word Formation
This section deals with the other basic domain of morphology, namely the
creation of new lexical units. It is important to note that more often than
not grammatical models tend to focus their interest on the semantic repre-
sentation of clauses and phrases, but they usually leave aside the question of
how to account for the semantic representation of other grammatical struc-
tures such as compound and derived words. Contrary to this tendency, the
model outlined in this section offers a design for the semantic representa-
tion of such units which complies with the following issues: firstly, it
captures the semantic variability of word-formation patterns. The introduc-
tion of a two-tier enriched representation of lexical structures, which
includes not only the event features encoded in logical structures, but also
encyclopedic knowledge as captured by Qualia Theory, places RRG in an
outstanding position when compared with other proposals of derivational
morphology. Section 8.4.1 deals with these aspects and Section 8.4.2 offers
some illustrative analyses of the most relevant affixal units in Spanish and
English. Secondly, such a format of semantic representation will allow us to
account for the grammatical relations that hold between the components of
a derived lexeme within its corresponding LSW. This will be the aim of
Section 8.4.3.
These two aspects reveal a conception of derivational morphology as a
double-sided phenomenon: as a lexical process, word-formation involves
the creation of a new semantic structure in the wrapper of a lexical
unit; at the same time, the constituents of a derivationally complex
lexical item are interrelated by different types of grammatical relations,
that is, word formation is also a syntactic phenomenon within the
word domain.

8.4.1 Word Formation as a Lexical Process: Lexical Representations and the

Derived Lexicon
As has already been mentioned, there is a great portion of morphology that
is lexical in nature. Word formation unquestionably falls into the realm of
lexical semantics, though the vast majority of models tend to overlook this
aspect. Levin and Rappaport Hovav (2001: 248–249) offer a brief but
revealing description of the neglect of the relationship between lexical
semantics and morphology in generative approaches, especially after Lie-
ber (1980). Even in those cases in which attention is paid to argument
structure operations in morphological theories (such as, for example, in
Baker 1988; Bresnan 1982; Rosen 1989 and Williams 1981, among many
others), such operations are essentially syntactic, not semantic. They also
cite a fundamental reason for this neglect, namely that these models often
lack a comprehensive theory of lexical-semantic representations, a funda-
mental framework for a sound study of such a feature of morphology.
A notable exception is the work of Lieber (2004), which can be considered a
major contribution to the lexical semantics of morphology. She contends
that a precondition for the study of derivational morphology is the exist-
ence of a solid theory of lexical-semantic representation. She also explains
that such a theory must be decompositional and based in semantic primi-
tives or atoms ‘of the right “grain size” to allow us to talk of the meanings
of complex words’ (Lieber 2004: 4); furthermore, semantic representations
must facilitate the analysis of lexical semantics and not only of the seman-
tic properties of higher syntactic structures (such as phrases or clauses).
Lastly, it must also allow us to describe the meanings of complex words in
the same terms as the meanings of simplex lexemes.
RRG can also claim to be exceptional with regard to this issue:
lexical-semantic representations lie at the heart of the model, and mor-
phosyntactic phenomena are always externally motivated by the semantic
representations assembled from the lexical-semantic structure of its
components. There have been many important contributions to the
enhancement of RRG lexical representations in the primary lexicon,14
and in the derived lexicon (in the works of Cortés-Rodríguez 2006a,
2006b, 2009; Cortés-Rodríguez and Sosa 2008, 2012, also Nolan 2011). From
all these studies it is possible to obtain a detailed picture of the overall
configuration of the lexicon in RRG and, specifically, of the structure of
lexical semantic representations.
As was stated in Section 8.2, the lexicon consists of lexical morphemes
of two types: free lexical morphemes or lexemes (which would belong to
the traditionally named open word classes) and bound lexemes (deriv-
ational affixes). They differ only in their distributional freedom of occur-
rence in discourse. Both the primary and the derived lexicons are
organized internally in lexical groupings of semantic and syntactic affin-
ity, which can be termed lexical classes (see Faber and Mairal Usón 1999
for an overview of the hierarchical organization of the English verbal

lexicon). The lexical entries for both types of lexeme include a formal
semantic representation in the format of lexical templates. The latest
version of lexical templates for free lexemes (Cortés-Rodríguez 2009;
Mairal Usón, Periñan-Pascual and Pérez Cabello de Alba 2012; Ruiz de
Mendoza 2013) takes as a backbone for lexical representation the Aktion-
sart properties of predicates as formally encoded in the logical structures
of RRG and enriches them by integrating Pustejovsky’s (1995) Qualia
Theory together with the set of lexical functions from the explanatory
and combinatorial lexicology (Mel’cuk 1989; Mel’cuk and Wanner 1996;
Alonso and Tutin 1996). Despite their complexity, lexical templates are
fully fledged repositories of the semantic features associated with a
lexeme, either free or bound. By way of example, the lexical template
for the Spanish verb captar ‘fathom’ would have the following format
(Cortés-Rodríguez and Sosa 2012: 22):
(10) captar:
eventstr: know′ (x, y)
qualiastr: {Qf: manner : MagnObstr think′ (x, y)
Qt: Culm know′ (x, y <all>)}
The format of a lexical entry now consists of two basic components: the
event and the qualia structure. In this case, since captar belongs to the class
of Cognition verbs, the event structure is a state logical structure which
takes know′ as a primitive and has two arguments (x, y), following the
format of the logical structures of RRG. The second part of the lexical
template includes its qualia structure. Qualia Theory is borrowed from
Pustejovsky’s (1995, 2001) semantic model. Van Valin and LaPolla (1997:
184–186) already incorporated qualia as part of the semantic representation
of nouns. Our proposal takes it beyond this and aims at including it as a part
of any lexical semantic representation. What qualia tell us about a concept
is the set of semantic constraints by which we understand a word when
embedded in the language, which is also the purpose of our lexical tem-
plates. As Pustejovsky and Batiukova (2019: 162) state:
The core notion of qualia structure is that there are four aspects that
make up our knowledge of a word: the class of entities it denotes (the
formal role or quale), how the denoted entity was created (the agentive
role), the intended function of this entity (the telic role), and the
internal makeup of this entity (the constitutive role).
They offer as an example the qualia structure of the noun violin.
(11) violin (x)

QS ¼ Formal ¼ musical_instrument (x)
Agentive ¼ build (y, x)
Telic ¼ produce_music_on (z, x)
Constitutive ¼ strings_of (w, x)

Turning to the representation of Spanish captar in (10), the template

encodes two qualia characterizations: the formal quale Qf encodes a
subevent in which a semantic function manner is combined with the
lexical functions ‘Magn’ (‘intensity’) and Obstr (‘difficulty’); the combin-
ation of these functions in turn modify the primitive think′. Thus, the
formal quale describes the degree of difficulty in carrying out the process
of thinking implicit in the meaning of captar. The telic quale as encoded
in Qt: Culm know′ (x, y) depicts the section of the event that expresses
the final aim of the event, to reach knowledge or to understand. As can
be seen, one of the goals of the qualia structure section is to represent
the semantic attributes by means of which a lexeme is semantically
distinguished within the larger set of units that belong to its
lexical class.
These two components of lexical entries, event structure and qualia, are
similar to the so-called ‘skeleton’ and ‘body’ of lexical representations in
Lieber’s (2004, 2009) proposal. However, we believe that the lexical entries
of RRG have some advantages over Lieber’s system: the event structure
captures in essence the semantic features encoded in the logical structures
of RRG. Logical structures are formalized representations of the Aktionsart
features of a given unit and are solid typologically based representations.
Even though Lieber’s skeleton is meant to include a set of universal
semantic features, no such set is described, and she defines only those that
are syntactically relevant for English (Lieber 2009: 83). Similarly, the body,
or set of semantic/pragmatic features which form encyclopedic knowledge,
is to a certain extent similar to qualia features, though Lieber allows for
some ‘unsystematicity’, insofar as a part of the ‘body’ (the ‘fat’; Lieber 2009:
83) varies from one individual to another. As regards the other part of the
‘body’ (the ‘muscle’), which is shared by a community, again a coherent
and systematic description of what it consists of is still lacking. In com-
parison, qualia theory is restrictively based on the Aristotelian modes of
explanation or aitia, meant to give a principled account of how humans
understand the world. As described in Pustejovsky and Batiukova (2019:
162), qualia structure is ‘a relational system whose parameters allow us to
decompose word meaning in a principled way, accounting for what other
concepts and words it can be associated with in different contexts, based
on its meaning’. Furthermore, qualia features and event structure descrip-
tions are formally represented by means of the same metalanguage, which
enhances the interaction between all elements in semantic representa-
tions of lexical and syntactic structures in RRG.
8.4.2 Some Illustrative Analyses

In a similar way to free lexical units, affixes are also clustered in affixal
classes, and their behaviour as members of these classes is also parallel to

that exhibited by words in the primary lexicon. Some units are more
central or higher in the structure whereas other affixes are peripheral,
and some are even located in between two lexical classes. In this section a
brief description of some of the most relevant affixal classes in Spanish and
English will be offered (for a more detailed account, see Cortés-Rodríguez
2006a, 2006b, Cortés-Rodríguez and Sosa 2008, 2012). More specifically, an
account of the semantic aspects of one important group of nominaliza-
tions is given first, followed by an explanation of the intricate meaning
relationships between the components of derived causative
verbal formations.
8.4.2.1 Macrorole Nominalizations

Let us consider how the semantic content of the Spanish nominalizing
affixes -ero (e.g. camión ‘lorry’ > camionero ‘lorry-driver’, guerra ‘war’ >
guerrero ‘warrior’, café ‘coffee’> cafetera ‘coffee pot’), -or (e.g. afilar ‘to
sharpen’ > afilador ‘sharpener’, calentar ‘to heat up’ > calentador ‘heater’,
despertar ‘to wake up’ > despertador ‘alarm clock’, poseer ‘to own’ > poseedor
‘owner’) [-a/-e/-ie]nte (e.g. humillar ‘to humiliate’ > humillante ‘humiliating’,
galopar ‘to gallop’ > galopante ‘rapidly advancing / out of control’, desinfectar
‘to disinfect’ > desinfectante ‘disinfectant’), -ista (e.g. novela ‘novel’> novelista
‘novelist’, ébano ‘ebony’ > ebanista ‘cabinetmaker’, izquierda ‘left’ > izquier-
dista ‘leftist’), [-a/-i]do (e.g. embravecer ‘to become rough’ > embravecido ‘furi-
ous/rough’, errar ‘to err’ > errado ‘mistaken/wrong’) and -ario (e.g. arrendar
‘to rent’ > arrendatario ‘tenant’, concesión ‘licence/concession’ > concesionario
‘licensee/franchisee’) and also English -er, -ist, [-a/e]-nt, -ee is represented in
this proposal:
(12) φiN: [(xi. . ., [(e2: [LT. . .φBASE: (Qualia:. . .xi)])])], where

x ¼ macrorole
φ ¼ lexical unit
LT ¼ lexical template
φBASE ¼ base lexeme
i ¼ denotational co-indexation
This formula reads as follows: Any derived nominal lexeme φiN belonging to
this class denotes a participant entity xi of one of the events encoded in the
qualia structure (Qualia:…xi) that forms part of the lexical template of the
base lexeme LT…φBASE; hence the derived lexeme and that entity are co-
indexed by a superscript i. Another way to express this is to describe these
processes as nominalizations oriented towards an argument in the semantic
representation of the lexical unit that functions as the base of the deriv-
ation. There is an additional restriction on this entity: it must have macro-
role status. Therefore, the structure in (12) represents the ‘event structure’ of
the derivational class of concrete nominalizing affixes, or concrete proces-
sual substances/things/essences in Lieber’s (2004: 36) terminology. We propose

labelling this derivational class as ‘macrorole nominalizations’ since they

encompass both actor and undergoer nominalizations. Actor nominaliza-
tions include derived instruments, agents, experiencers, locations, etc., and
undergoer formations are prototypically formed in Spanish by means of
(-a/-i)do and -ario and in English by -ee and -ed. The two corresponding
templates are the following.
(13) φiN: [LT (xi. . ., [(e2: [LT. . .φBASE: (Qualia:. . .xi)])])], x ¼ actor
E.g. escritor ‘writer’, poseedor ‘owner’, panfletista ‘pamphleteer’, desodorante
‘deodorant’, lechero ‘milkman’.
(14) φiN: [LT (. . .xi, [(e2: [LT. . .φBASE: (Qualia:. . .xi)])])], x ¼ undergoer
E.g. arrendatario ‘tenant’, fideicomisario ‘trustee’, fallecido ‘deceased’
The labels ‘actor and undergoer nominalizations’ explain the wide scope
of these types of derivational processes. The semantics of the templates
cannot be associated with a specific semantic function such as ‘agent’.
Even though the most prototypical formations correspond to agent nom-
inals, like English writer, runner, violinist, etc., there are many other forma-
tions where the notion of ‘agenthood’ is absent (cf. formations like believer,
owner, lover, to mention just a few). The term ‘actor’ indicates that all the
formations are nominalizations of the macrorole actor, as defined within
RRG (see Chapter 4). This, in turn, justifies the superscript i which co-
indexes the lexical variable for the derived word (φN) with the participant
that would receive that macrorole function. That is, they mark the nom-
inals as oriented towards one entity (xi) involved in one of the subevents
that form part of the state of affairs depicted by the base word. The
variable LT (‘lexical template’) indicates that the event in which this entity
participates can be of any kind, a state (pred′), an activity (do′), or any
other logical structure.
The following representation is to be understood as a subspecification of
the actor template, and corresponds to the traditionally labelled ‘agent
nominals’, which in RRG terms should be described as ‘effector nominaliza-
tions’ (Van Valin and Wilkins 1996).
(15) φiN: [do′ (xi, [φBASE])], x ¼ actor. E.g. driver, runner, smoker.
This structure expresses the semantic content of the most prototypical nom-
inalizations within the class: the derived words corresponding to this con-
struction describe the effector involved in the event expressed by the
semantics of the base word. There are two co-indexing possibilities expressed
in the above representation depending on whether the formation is deverbal
(φV) or not (φ[-V] ). In the case of deverbal effector nouns, co-indexation is
usually quite straightforward: the verbal bases typically encode an event that
is dynamic, and therefore the meaning of effectorhood derives from the
semantic function of its first argument. This is the case of hunter.

(16) hunter
[X-er] N: [ (xi , [huntv ])], x = actor
huntV: eventstr: [do′ (x, e1)]
OUTPUT hunter iN : argstr: x: animate

qualiastr: {…QA: e1 [do′ (x i , [hunt′(x,y)])]}
The semantic structure of the Spanish nominal creyente ‘believer’ in (17)

illustrates the fact that these kinds of derived actor nominals do not
exclusively refer to agent arguments. Let us recall that the actor macrorole
is assigned to the leftmost argument in a logical structure with two
arguments, irrespective of the type of event encoded (Van Valin 2005:
60–67).
(17) creyente ‘believer’

[X(-a/-e/-ie)nte] N: [(xi , [(creerV)])], x = actor
creerV: eventstr: e1[pred′ (x,y)]
OUTPUT creyenteiN: argstr: x: human

qualiastr: {…QA: e1[believe' (xi,y)]}
The semantic interpretation of denominal actor lexemes – such as pianista

‘pianist’, cabrero <cabra ‘goat’þero> ‘goatherd’, banquero ‘banker’, pensionista
‘pensioner’ – likewise arises from co-indexation with one argument in one of
the events depicted in the qualia structure of the base noun, as is the case of
violinista, where semantic composition is based on the event encoded as the
telic quale of violin.
(18) violinista ‘violinist’

[violinN + istaiN] iN: [DO (xi ACTOR , [do′ (xi, [play′ (xi, violinN)])])]
pianoN: typestr = (y: artifact-lcp)

argstr: y: phys_object
qualiastr: {Qformal: musical.instrument (y)
Qconst.: pegs_strings_bow_....of (w,y)
Qagent.: e1[do′ (w, [create′ (z,y)])]
Qtelic: e1[do′ (xi, [play′(xi, y)])]}

The representation in (18) also reveals that there are interesting semantic
differences among the members of this subclass. The affix -ista is more
restrictive than other members among the actor nominalizing affixes in
terms of: (a) the types of bases (it prefers nouns and adjectives, in contradis-
tinction with the more deverbal -or and -nte); and (b) its meaning, as it always
involves a degree of volitionality not necessarily present in the other affixes.
Thus, its LT would be more specific or hyponymic.
(19) [φBASE: N/adj þ istaiN] iN : [DO (xi, [do′ (xi, [e2: LT . . .φBASE: (Qualia: . . .LT (xi…)])])]
(19) is a good instance of the kind of information that is encoded in an

affixal LT. It includes a morphological frame with the constituent makeup
of the derived words: [φBASE:N/adj þ istiN] iN (cf. Nolan 2011). This frame
includes information of the categories of both the base [φBASE:N/Adj] and the
complex word (. . .] iN).
The semantic specificity of -ista is expressed through the agent operator
([DO (xi,) (see Van Valin and LaPolla 1997: 118–120), which is understood as
an overlay of the effector thematic role. Note that the specific semantic
content of this formation is spelled out by the joint effect of the selection of
one quale from the LT of the base word, and co-indexation of the affix with
one of the arguments involved in the subevent of such a quale.15
Macrorole nominalizations are probably the most significant class to look at
if we are to fully appreciate how relevant co-indexation is for the construction
of the meaning of complex words. However, there is another mechanism that
is to be used in this meaning-construction process: qualia selection. Qualia
selection is essential to understand the semantics of derived causative verbs.
8.4.2.2 Derived Causative Verbs

One of the great difficulties of a lexical-semantic approach to word formation
is the rampant polysemy that most affixes and composition patterns exhibit.
Jackendoff (2009: 117) explains that a compound like box car can have a
plethora of paraphrases (car that carries boxes, that looks like a box, that is used as
a box, etc.) However, this does not mean that complex words are ambiguous or
vague; Jackendoff describes them as semantically ‘promiscuous’, which means
that it is possible for a complex word to take up all those meanings. A similar
approach is defended in the works on word formation within Pustejovsky’s
Generative Lexicon framework (Batiukova 2008; Johnston and Busa 1999). This
proposal follows the same path in considering that complex words are prima
facie semantically and grammatically underspecified structures. However, this
does not mean that there are no limits to compositionality. All possible
interpretations of a complex word must be anchored – directly or indirectly –
in the richer semantic information encoded in the LTs of its components, and
all word-formation processes involve the activation of mechanisms to select
the meaning or network of possible meanings of the complex word. In this
regard, qualia selection from LTs plays a fundamental role.

The derivational class of causative verbal lexemes will be of help to

provide examples of this phenomenon. The basic lexical template for this
class is as follows.
(20) i
[ BASE + V ] V: [do′ (x, Ø )] CAUSE [ e2 : (LT : … BASE ... )]
base <Quale i >

= lexical unit
LT = lexical template
BASE = base lexeme
i = denotational co-indexation
The formula in (20) encodes a complex semantic structure in which there is a

causal bond between two subevents, the induced one corresponding to a state
of affairs in which the base word is involved or affected more or less directly.
Typically, causative formations are category-changing processes, the bases
being nouns or adjectives. Some conversion processes (e.g. conciencia ‘aware-
ness’ > concienciar ‘raise awareness’, completo ‘complete’ > completar ‘to com-
plete’, alegre ‘happy’ > alegrar ‘gladden’) and the affixes a- (e.g. feo ‘ugly’ > afear
‘spoil, make ugly’, largo ‘long’ > alargar ‘lengthen’), -iz- (e.g. suave ‘soft’ >
suavizar ‘soften’, colonia ‘colony’ > colonizar ‘colonize’), -ific- (e.g. sólido ‘solid’
> solidificar ‘solidify’, puro ‘pure’ > purificar ‘purify’), and en- (e.g. ancho ‘wide’ >
ensanchar ‘widen’, lata ‘can’ > enlatar ‘to can’, jaula ‘cage’ > enjaular ‘encage’)
are members of this derivational class in Spanish.
The semantic interpretation of every complex word that is an output of
causativization is determined by the selective binding or exploitation of one
quale of the LT of the base, expressed in (20) by φBASE<Quale i> (cf. Cortés-
Rodríguez 2006a; Batiukova 2008). To be more precise, the quale that is
selected for binding will endow the caused subevent in (20) ([e2:
(LT: . . .φBASE. . . )]) with a specific content. For instance, a formation like Spanish
enlatar ‘to can’ has a causative-locative interpretation, which is promoted by
the selection of the formal quale (Qf: container′ (xi, y)) of the noun lata ‘can’.16
Consequently, the abstract operator LT corresponding to the caused subevent
would receive a specific locative interpretation (BECOME be-in′ (xi, y)]); co-
indexation (represented by the arrow in (21)) specifies further that the base
entity designates a locus (Cortés-Rodríguez and Sosa 2012: 27).
(21) enlatar (en ‘cause’ + lata ‘can’ enlatar ‘to can’)

[en- + lataNi ]V: do′ (z, Ø) CAUSE [e2= [BECOME be-in′ (lataiN , y)]]
TYPESTR = (x: artifact-lcp)
lataN QUALIASTR= … QF: container′ (xi,y)

QC: metallic′ (x)
QA: e1[do′ (z, Ø)]/ artifact′ (x y)
QT: e2: [BECOMEbe-in′(xi, y)]

On the other hand, deadjectival forms like Spanish suavizar ‘soften’, ensan-
char ‘widen’, solidificar ‘solidify’, etc. select the formal quale of the base and
the meaning of the complex word will be that of a prototypical causative-
resultative verb (‘cause become adj’), as shown in (22) (Cortés-Rodríguez and
Sosa 2012: 28).
(22) suavizar (suave ‘soft’ + izar ‘cause’ suavizar ‘to soften’)

[suaveAdj i + izar V ]V : do′ (z, Ø ) CAUSE [e 2 = [BECOME suave′ (y)]]
suaveAdj i : {…QF: be′(y, suave′i ) …}
As can be seen, co-indexation works hand-in-hand with qualia selection:

its function is to bind one element from the base template with another
from the affixal template, or the modifying lexeme in the case of
compounds.17
8.4.3 Word Formation as a Grammatical Process: The Interface between

Lexical Templates and the LSW
We are now in a position to consider how the semantic structures under-
lying word-formation products of the kind outlined previously can find a
syntactic counterpart in a syntactic theory of the word. Martín Arista’s
(2008, 2009, 2011) LSW as described in Section 8.2 offers a number of
significant advantages for this task: it will enable us to design a semantics-
to-syntax interface between the semantic descriptions proposed here and
the layered structure of the corresponding complex word. In fact, only a few
adjustments are needed in order to achieve this goal. The most important of
these would be to reassess the grammatical status of some constituents in
the LSW, as proposed in the original model. A good example for this would
be the LSW of bookseller, analysed in Martín Arista (2009: 92) as an exocentric
formation (Figure 8.10).
This analysis, which takes as nucleus the verbal base, seems to be based on
the fact that the elements of word syntax are a kind of carbon-copy of what
they could be at clause level; this is what seems to be stated in the following
quotation (Martín Arista 2009: 94):
The definition of Word functions is based on Clause functions,

which requires an indirect association with a clausal expression:
incomer expresses a First Argument whereas income expresses a
Second Argument, incomer, outflow and inflow express an Argument-
Adjunct, etc.
However, if we want to do justice to the lexical-semantic dimension of

word formation, we must provide a semantic motivation for the functional
constituents of the LSW, which in turn would be more in line with what is
done in the other RRG grammatical domains (Van Valin and LaPolla
1997: 68):

The elements functioning as nuclei of PPs and clauses are predicates in

the semantic representation, while nominal nuclei are designated
within the semantic representation of the NP. Thus the head of a phrase
is a function of its semantics: an NP is headed by a nominal nucleus, a
PP by an adpositional nucleus, and a clause by a predicating nucleus.
COMPLEX WORDN
COREN
ARG NUCV ARGN
WORD PREDICATEV WORDN
CORE CORE
NUC NUCN
book sell er
Figure 8.10 Exocentric analysis of compound bookseller (Martín Arista 2009: 92)
The logical follow-on from this statement is that if a word has a referential
semantic function, its nucleus (or head) should be the element that confers
such a value. We can borrow some criteria from Štekauer (2005b: 225–226) to
select the nucleus element within the LSW of a complex word: (i) Hyponymy:
the complex word is a semantic specification of the head; storyteller, for
example, is a hyponym of teller, which in turn is a hyponym of -er. (ii) Subcat-
egorization: heads impose subcategorization restrictions; -ist selects noun
bases, -en has a strong preference for monosyllabic bases with final plosives,
as in red > redden, short > shorten. (iii) Distributional and categorial equiva-
lence: the head (or nucleus) specifies the lexical class of the derived word.
For example, -er formations are consistently denominal; on the other hand,
locative prefixes (pre-field or sub-way) are not heads as they do not determine
the class form of the derived word.
This last criterion draws a key parallel between the nature of the head or
nucleus of clauses and phrases, on the one hand, and complex words, on the
other. If we take these conditions into account, the layered structure of book-
seller could be as shown in Figure 8.11 (Cortés-Rodríguez and Sosa 2012: 36).
The structure captures the endocentric character of the bound lexeme -er
in the derivation of seller and the recursive compositional process with book.
Figure 8.12 shows the structure of non-endocentric complex words in

COMPLEX WORDN
COREN
NUCLEUSN
WORDN PREDICATEN
COMPLEX WORDN
CORE N COREN
NUCN NUCN
PREDICATEN WORDV PREDICATEN
COREV
NUCV
PREDICATEV
book sell er
Figure 8.11 Endocentric analysis of compound bookseller (Cortés-Rodríguez and Sosa 2012: 36)
Spanish, which includes compounds like lavaplatos ‘dishwasher’, sacacorchos

‘corkscrew’, pagafantas (lit. ‘Fanta payer’) ‘conned cuckold’ and conversion
formations like destino ‘destiny’, ‘end use ’> destinar ‘assign’, amigo ‘friend’ >
amigar ‘make/become friends’, deseo ‘desire’ > desear ‘to desire’. Co-
indexation in all these cases is revealing: none of the components of the
morphological template is co-indexed with the lexical variable of the com-
plex word [lavaN/adj þ platosNα] iNβ, in sharp contrast with endocentric
formations as in pianista [pianoN þ istaiN] iN. Thus, it may be more appropri-
ate to describe these complex words as ‘acentric’, since they do not have a
lexically saturated nucleus (Cortés-Rodríguez and Sosa 2012: 37).
8.5 Conclusion
The aim of this chapter was to present an overview of morphology in

RRG. As stated at the outset, despite the obvious recognition of the
centrality of morphological structure, there is as yet no fully articulated
theory of the structure of the word domain and the way it interacts with

COMPLEX WORDN
COREN
WORDV WORDN NUCN
COREV COREN
NUCV NUCN
PREDV PREDN
lava platos
[lavaN/ADJ + platosNα] iNβ : [do′ (xi , [ washV-BASE ]), x= Actor
washV : EVENTSTR: e1 < e2 / e1 CAUSE e2

QFORMAL : (e2: [do′ (xi , [wash'(xi, platos)])
QAGENT: (e1: [do′ (y, [use′ (y, xi)])])
Figure 8.12 ‘Acentric’ Spanish compound lavaplatos (‘dishwasher’) (Cortés-Rodríguez and Sosa
2012: 37)
the structure of higher grammatical domains such as phrases and

clauses. However, there are some significant contributions leading to
the development of such a morphological theory. The first challenge,
the establishment of descriptive tools for the structure of the word, was
the topic of Section 8.2, in which we offered the guidelines for a layered
structure of the word. Given the explanatory character of the model and
its functional (or better, communication-and-cognition) orientation,
Sections 8.3 and 8.4 were dedicated to accounting for the functional
motivation of inflection and word formation, respectively. In Section 8.3
we offered a description of the interface of inflection within and outside
the word. This has in turn revealed the potential of the LSW for the
explanation of the interaction between syntactic and semantic structures,
closely tied to the notion of ‘operator percolation’ between the LSW and
the LS of higher grammatical structures. However, the intricate relation
between word structure and clause structure goes beyond feature perco-
lation in the case of head-marking languages, as we also pointed out in
this section.
Section 8.4 was devoted to the explanation of word formation within
RRG. This involves identifying the nature of the connections that hold

between morphology, on the one hand, and lexical semantics and syntax,
on the other. As a lexicological phenomenon, derivational affixes and
processes form part of the lexicon in RRG and – very much like free
lexemes – are endowed with a semantic representation in the format of a
lexical template. The interaction of affixal and word templates is
accounted for by means of two explanatory devices: co-indexation and
qualia specification. The section closed with a brief overview of how the
semantic structure underlying a derivationally complex word also finds its
syntactic counterpart in the LSW, as described in Section 8.2, with some
slight adjustments to the proposal put forward by Martín Arista (2008,
2009, 2011).
In our opinion, taken together, the proposals outlined in this chapter,
which draw on several contributions both from inside and outside RRG,
constitute an explanatory framework for the study of morphological phe-
nomena within this model.
References
Ackema, P. and A. Neeleman. 2003. Syntactic atomicity. Journal of Comparative

Germanic Linguistics 6: 93–128.
Alonso Ramos, M. and A. Tutin. 1996. A classification and description of
lexical functions for the analysis of their combinations. In L. Wanner (ed.),
Lexical Functions in Lexicography and Natural Language Processing, 146–167.
Anderson, S. 1992. A-Morphous Morphology. Cambridge: Cambridge University
Press.
Aronoff, M. and K. Fudeman. 2011. What Is Morphology? (2nd ed.). Oxford:
Wiley-Blackwell.
Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing.
Baker, M. 2003. Lexical Categories. Cambridge: Cambridge University Press.
Batiukova, O. 2008. Morfología: del léxico a la sintaxis oracional. In Actas del
VIII Congreso de Lingüística General. Universidad Autónoma de Madrid.
www.lllf.uam.es/clg8/actas/index.html.
Bentley, D. 2018. Monotonicity in word formation: The case of Italo-
Romance result-state adjectives. Transactions of the Philological Society 116
(3): 285–319.
Bentley, D. 2019. The logical structure of verbs of quantized and non-
quantized change. Paper delivered at the 2019 International Conference on
Role and Reference Grammar. The State University of New York, University at
Buffalo (SUNY).
Bosque, I. 2012. On the lexical integrity hypothesis and its (in)accurate
predictions. Iberia: An International Journal of Theoretical Linguistics 4(1):
140–173.

Boutin, M. 2011. Towards a realizational approach to morphology in RRG. In

W. Nakamura (ed.), New Perspectives in Role and Reference Grammar, 2–34.
Bresnan, J. (ed.). 1982. The Mental Representation of Grammatical Relations.
Cambridge, MA: MIT Press.
Bybee, J. 1985. Morphology: A Study of the Relation between Meaning and Form.
Cortés-Rodríguez, F. J. 2006a. Derivational morphology in Role and Reference
Grammar: A new proposal. RESLA: Revista Española de Lingüística Aplicada 19:
41–66.
Cortés-Rodríguez, F. J. 2006b. Negative affixation within the Lexical Gram-
mar Model. RÆL: Revista Electrónica de Lingüística Aplicada 5: 27–56.
Cortés-Rodríguez, F. J. 2009. The inchoative construction: Semantic
representation and unification constraints. In C. S. Butler and J. Martín
Arista (eds.), Deconstructing Constructions, 247–270. Amsterdam: John
Benjamins.
Cortés-Rodríguez, F. J. and E. Sosa. 2008. The morphology–semantics inter-
face in word-formation. Revista Canaria de Estudios Ingleses 57: 91–108.
Cortés-Rodríguez, F. J. and E. Sosa. 2012. La morfología derivativa. In R.
Mairal Usón, L. Guerrero and C. González (eds.), El funcionalismo en la teoría
lingüística: La Gramática del Papel y la Referencia, 19–42. Madrid: Akal.
Di Sciullo, A. M. and E. Williams. 1987. On the Definition of Word. Cambridge,
MA: MIT Press.
Diedrichsen, E. 2011. The layered structure of the German word: An RRG
approach to inflectional morphology. Ms. Paper delivered at the 2011
International Course and Conference on Role and Reference Grammar. The Ponti-
ficia Universidad Católica de Chile, Santiago, Chile.
Diedrichsen, E. and B. Nolan. 2011. The syllable in the LSW: A proposal for an
RRG morphophonological interface. Ms. paper delivered at the 2011 Inter-
national Course and Conference on Role and Reference Grammar. The Pontificia
Universidad Católica de Chile, Santiago, Chile.
Dik, S. C. 1997. The Theory of Functional Grammar. 2 vols. Berlin: Mouton de
Gruyter.
Everett, D. 2002. Towards an RRG theory of morphology. Lecture given at the
2002 International Conference on Role and Reference Grammar, Universidad de
La Rioja, Logroño, Spain.
Faber, P. and R. Mairal Usón. 1999. Constructing a Lexicon of English Verbs.
Foley, W. A. and R. D. Van Valin, Jr. 1984. Functional Syntax and Universal
Grammar. Cambridge: Cambridge University Press.
Hengeveld, K. 1989. Layers and operators in Functional Grammar. Journal of
Jackendoff, R. 2009. Compounding in the Parallel Architecture and concep-
tual semantics. In R. Lieber and P. Štekauer (eds.), The Oxford Handbook of
Compounding, 105–128. Oxford: Oxford University Press.

Johnston, M. and F. Busa. 1999. Qualia structure and the compositional

interpretation of compounds. In E. Viegas (ed.), Breadth and Depth of Seman-
tic Lexicons, 167–187. Dordrecht: Kluwer Academic.
Lapointe, S. 1981. A lexical analysis of the English auxiliary verb system. In T.
Hoekstra, H. van der Hulst and M. Moortgat (eds.), Lexical Grammar,
215–254. Dordrecht: Foris.
Levin, B. and Rappaport Hovav, M 2001. Morphology and lexical semantics.
In A. Spencer and A. M. Zwicky (eds.), The Handbook of Morphology,
pp. 248–271. Oxford: Blackwell.
Lieber, R. 1980. On the Organization of the Lexicon. PhD dissertation, MIT
(published by Garland Press, New York, 1994).
Lieber, R. 2004. Morphology and Lexical Semantics. Cambridge: Cambridge
University Press.
Lieber, R. 2009. A lexical semantic approach to compounding. In R. Lieber
and P. Štekauer (eds.), The Oxford Handbook of Compounding, 78–104. Oxford:
Mairal Usón, R. and Cortés-Rodríguez, F. J. 2000–2001. Semantic packaging
and syntactic projections in word formation processes: The case of agent
nominalizations. RESLA (Revista Española de Lingüística Aplicada) 14:
271–294.
Mairal Usón, R. and F. J. Ruiz de Mendoza. 2008. New challenges for lexical
representation within the Lexical-Constructional Model. Revista Canaria de
Estudios Ingleses 57: 137–158.
Mairal Usón, R. and F. J. Ruiz de Mendoza. 2009. Levels of description
and explanation in meaning construction. In C. S. Butler and J. Martín
Arista (eds.), Deconstructing Constructions, 153–198. Amsterdam: John
Benjamins.
Mairal Usón, R., C. Periñán-Pascual and M. B. Pérez Cabello de Alba. 2012. La
representación léxica: Hacia un enfoque ontológico. In R. Mairal Usón, L.
Guerrero and C. González (eds.), El funcionalismo en la teoría lingüística: La
Gramática del Papel y la Referencia, 85–102. Madrid: Akal.
Marchand, H. 1969. The Categories and Types of Present-Day English Word Forma-
tion, 2nd ed. Munich: Beck.
Martín Arista, J. 2008. Unification and separation in a functional theory
of morphology. In R. D. Van Valin, Jr. (ed.), Investigations of the Syntax–
Semantics–Pragmatics Interface, 85–115. Amsterdam: John Benjamins.
Martín Arista, J. 2009. A typology of morphological constructions, In
C. S. Butler and J. Martín Arista (eds.), Deconstructing Constructions, 85–116.
Amsterdam: John Benjamins
Martín Arista, J. 2011. Projections and constructions in functional morph-
ology: The case of Old English HRĒOW. Language and Linguistics 12(2):
393–425.
Martín Arista, J. 2012. La morfología flexiva. In R. Mairal Usón, L. Guerrero
and C. González (eds.), El funcionalismo en la teoría lingüística: La Gramática
del Papel y la Referencia, 43–58. Madrid: Akal.

Martín Mingorance, L. 1998 [1985]. Bases metodológicas para un estudio

contrastivo del léxico derivado. In A. Marín Rubiales (ed.), El Modelo Lex-
emático Funcional, 61–82. Granada: Editorial Universidad de Granada.
Mel’cuk, I. 1989. Semantic primitives from the viewpoint of the Meaning-
Text Linguistic Theory. Quaderni di Semantica 10(1): 65–102.
Mel’cuk, I. and L. Wanner. 1996. Lexical functions and lexical inheritance for
emotion lexemes in German. In L. Wanner (ed.), Recent Trends in Meaning-
Text Theory, 209–227. Amsterdam: John Benjamins.
Möllemann, R. 2016. Implications of German Word Formation Processes for a Role
and Reference Grammar Approach to Morphology. MSc, Heinrich Heine Univer-
sity Düsseldorf. www.semanticscholar.org/paper/Implications-of-German-
word-formation-processes-for-Mollemann/
fcc1cf700f876136c0300269f5391671155c3be1.
Nolan, B. 2011. Meaning construction and grammatical inflection in the
Layered Structure of the Modern Irish word: An RRG account of morpho-
logical constructions. In W. Nakamura (ed.), New Perspectives in Role and
Reference Grammar, 64–101. Newcastle upon Tyne: Cambridge Scholars
Publishing.
Nübling, D. 2008. Historische Sprachwissenschaft des Deutschen (2nd ed.). Tübin-
gen: Narr.
Pavey, E. L. 2010. The Structure of Language: An Introduction to Grammatical
Periñán-Pascual, C. and R. Mairal Usón. 2009. Bringing Role and Reference
Grammar to natural language understanding. Procesamiento del Lenguaje
Natural 43: 265–273.
Pustejovsky, J. 1995. The Generative Lexicon. Cambridge, MA: MIT Press.
Pustejovsky, J. et al. 2001. Advances in Generative Lexicon Theory. Dordrecht:
Springer.
Pustejovsky, J. and O. Baitukova. 2019. The Lexicon. Cambridge: Cambridge
University Press.
Rijkhoff, J. 2002. The Noun Phrase. Oxford: Oxford University Press.
Roberts, J. R. 2015. Distributives in Amele. A Role and Reference Grammar
analysis. SIL Electronic Working Papers 2015–001. www.sil.org/system/files/
reapdata/16/20/84/162084215824434557688405164453437430340/
silewp2015_001.pdf.
Roberts, T. 1999. Grammatical relations and ergativity in Sta’át’imcets (Lil-
looet Salish). International Journal of American Linguistics 65: 275–302.
Rosen, S. T. 1989. Two types of noun incorporation: A lexical analysis.
Language 65: 294–317.
Ruiz de Mendoza, F. J. 2013. Meaning construction, meaning interpretation
and formal expression in the Lexical Constructional Model. In B. Nolan
and E. Diedrichsen (eds.), Linking Constructions into Functional Linguistics,
Ruiz de Mendoza, F. J. and R. Mairal Usón. 2007a. High-level metaphor and
metonymy in meaning construction. In G. Radden, K. Köpcke, T. Berg and

P. Siemund (eds.), Aspects of Meaning Construction, 33–51. Amsterdam: John

Benjamins.
Ruiz de Mendoza, F. J. and R. Mairal Usón. 2007b. Levels of semantic repre-
sentation: Where lexicon and grammar meet. Interlingüística 17: 26–47.
Selkirk, E. O. 1982. The Syntax of Words. Cambridge, MA: MIT Press.
Spencer, A. 1998. Morphophonological operations. In A. Spencer and A. M.
Zwicky (eds.), The Handbook of Morphology, 123–143. Oxford: Blackwell.
Spencer, A. 2004. Morphology: An overview of central concepts. In L. Sadler
and A. Spencer (eds.), Projecting Morphology, 67–109. Stanford, CA: CSLI
Publications.
Štekauer, P. 2005a. Meaning Predictability in Word Formation. Amsterdam: John
Benjamins.
Štekauer, P. 2005b. Onomasiological approach to word-formation. In P.
Štekauer and R. Lieber (eds.), Handbook of Word-Formation, 207–232. Dor-
drecht: Springer.
Stump, G. 2001. Inflectional Morphology: A Theory of Paradigm Structure. Cam-
Van Valin, R. D., Jr. 2005. The Syntax–Semantics–Pragmatics Interface: An Intro-
duction to Role and Reference Grammar. Cambridge: Cambridge University
Press.
Van Valin, R. D., Jr. 2008. RPs and the nature of lexical and syntactic
categories in Role and Reference Grammar. In R. D. Van Valin, Jr. (ed.),
Investigations of the Syntax–Semantics–Pragmatics Interface, 161–178. Amster-
Van Valin, R. D., Jr. 2013. Head-marking languages and linguistic theory.
In B. Bickel, L. A. Grenoble, D. A. Peterson and A. Timberlake (eds.),
Language Typology and Historical Contingency, 91–124. Amsterdam: John
Benjamins.
Van Valin, R. D., Jr. and D. P. Wilkins. 1996. The case for ‘effector’: Case roles,
agents, and agency revisited. In M. Shibatani and S. Thompson (eds.),
Grammatical Constructions, 289–322. Oxford: Clarendon Press.
Van Valin, R. D., Jr. and R. LaPolla. 1997. Syntax: Structure, Meaning and
Function. Cambridge: Cambridge University Press.
Williams, E. 1981. On the notions ‘lexically related’ and ‘head of a word’.
Notes
1 We may even include phonology here, though this remains a very under-
developed area. In Diedrichsen and Nolan (2011) there is an initial contri-
bution in this field within RRG, and Möllemann (2016: 30–34) provides a
brief description of the basic conditions necessary for a morphology–
phonology interface in RRG.

2 In incremental theories it is assumed that lexical units acquire morpho-

syntactic features only after inflectional morphs are acquired; by way of
example, works has the features ‘3rd person singular, present, indicative’
only after the suffix -s is attached to it.
3 See Mairal and Cortés-Rodríguez 2000–2001, Everett 2002, Cortés-Rodrí-
guez 2006a, Martín Arista 2008, 2009 and Nolan 2011, for more details on
all these resources.
4 Van Valin (2005: 161) offers a similar overview of the lexicon, when he
states that ‘[i]t is necessary to think of the lexicon as having at least two
parts, one the traditional storehouse of words and morphemes, and the
second a “workshop” where lexical rules and other lexical processes create
new lexical forms which would not otherwise be stored’.
5 There are some arguments against morpheme-based approaches to inflec-
tion and the lexicon among which the following are worth mentioning,
albeit briefly. A morpheme-based approach considers morphemes as a linear
sequence of phonemes which are attached to a base. However, morphosyn-
tactic information can be encoded by other means such as tone, stress or
nasalization across languages. Further problems for concatenative
approaches occur when there is no one-to-one correspondence between
meaning and form. While morpheme-based approaches assume that a
morphological property has one exponent per word, realizational theories
have no such requirement and are thus seemingly closer to the different
possibilities found across languages. A case in point is portmanteau affixes
like the Finnish nominal -t, which bears two concurrent features, plural
number and nominative case (as in talot ‘houses’). In a linear item-and-
arrangement approach it would be necessary to subcategorize the nomina-
tive plural form as exceptional so that the grammar would not first select
the plural suffix and then the case suffix in accordance with the usual
morphotactics of the language. In a realizational approach -t is the cumula-
tive exponent of two simultaneous features triggered by the corresponding
realizational rule(s) (see Boutin 2011: 5–7). Again, realizational rules will
account for this phenomenon in the morphology–phonology interface.
6 It is precisely here that the main difference between Martín Arista’s (2009,
2011) description of the LSW and Everett’s (2002) initial proposal lies:
whereas Everett (2002) places inflection in the constituent projection of
the word as daughters of the core layer (see Van Valin 2013: 112), Martín
Arista (2009: 90) proposes instead to treat inflection as part of the operator
projection, in which inflectional features are attached at the different
layers within the LSW.
7 Their status as arguments is expressed by the nodes ARG in Figure 8.3.
Although ARG nodes have been dispensed with in the LSC, Martín Arista
(2011) makes use of them in the analyses of morphologically complex
words. We have also opted to keep them in order to facilitate the inter-
pretation for those readers who may not be familiar with syntactic ana-
lyses within the word.

8 Diedrichsen (2011: 5–10) offers a similar strategy for the distribution of

operators in the LSW of German nouns and verbs. Following Bybee’s
(1985) principle of relevance, she correlates the degree of formal fusion
of an inflectional category and its bases with the semantic influence of
such a category. According to this, the more relevant a category, the
closer its degree of fusion or attachment with its base. Number, for
example, has a strong impact on nouns, as it involves a multiplication
of the concept it expresses. Because of this, stem modulation processes
such as umlaut for number marking in German nouns (Lamm ‘lamb’/
Lämm-er ‘lamb-s’; Buch ‘book’/ Büch-er ‘book-s’) will be interpreted as a
realization of a nuclear operator; on the other hand, number is less
relevant within the verbal inflectional categories and concomitantly it
will be encoded in the outer word layer.
9 Compare Russian atelic verbs like pisat’ ‘to write’ and pit’ to drink’, which
are morphologically simple words, with telic na-pisat’ ‘to write (up)’, pod-
pisat’ ‘to sign’ or vypit’ ‘to drink (up)’, all prefixed. Van Valin and LaPolla
(1997: 90–112) offer a plethora of examples of this type of process from a
wide array of languages.
10 See Roberts (2015) for a detailed description of the system of distributives
in this language.
11 In the glosses of example (3) ds stands for ‘different subject following’
and su for ‘subject’.
12 See Van Valin and LaPolla (1997: 3–7) for a detailed description of (theory
and language) internal and external explanations.
13 This view goes against the Lexical Integrity Hypothesis (see Lapointe 1981;
Di Sciullo and Williams 1987; Anderson 1992, etc.), which stipulates that
morphology is not visible to the rules of syntax. For arguments in favour
of and against this hypothesis, see Ackema and Neeleman (2003) and
Bosque (2012), respectively. In clear contrast to formalist modular
approaches, in RRG all elements of the theory interact with one another,
although their interaction is subject to various constraints. If one assumes
Everett’s (2002) version of the LSW, as Van Valin (2013) does, only inflec-
tional features of the COREw are visible to syntax, whereas the internal
structure of the NUCLEUSw is not visible to syntax. So in nouns like cats
and doorstops the syntax can ‘see’ the plural inflection, as it is relevant for
agreement phenomena, but it cannot see that cats has a single element in
its NUCLEUSw since the plural suffix -s is an argument in the COREw, but
doorstops has a complex nucleus composed of two compounded NUCLEIw.
In Martin Arista’s (2008, 2009, 2011) version of the LSW, in which inflec-
tional features are treated as operators, syntax-blind features are analysed
as NUCLEUSw operators. On the other hand, COREw and WORDw oper-
ators are visible to syntax and can access the LSC and the LSW, as shown in
Figures 8.5, 8.6 and 8.9.
14 Among them are the works of Mairal Usón and Ruiz de Mendoza (2008,
2009), Mairal Usón, Periñan-Pascual and Pérez Cabello de Alba (2012),

Ruiz de Mendoza (2013), Ruiz de Mendoza and Mairal (2007a, 2007b),

Periñán-Pascual and Mairal Usón (2009) and Bentley (2018, 2019).
15 Möllemann (2016: 12–15) proposes a means to establish the meaning of a
morpheme: it must be the semantic material left after removing the host
word’s contribution to the semantics of the derived word. Since seman-
tic information is encoded as event and qualia structure, the implication
is that an affixal entry will have an incomplete part in the qualia set, to
be filled once it is combined with a host lexeme.
16 Other formations of this type are acorralar (corral ‘pen, corral’ > a.corral.
inf ‘to corner, to corral’), entronizar (trono ‘throne’ > en.tron(o).iz.inf
‘enthrone’), encarcelar (cárcel ‘jail’ >en.carcel.inf ‘to jail’), alunizar (luna
‘moon’ > a.lun(a).iz.inf ‘to land on the moon’).
17 See Johnston and Busa (1999) for a very similar analysis to NþN com-
pounds within the framework of the Generative Lexicon theory.

9
Adverbs, Mimetics
and Ideophones
Kiyoko Toratani
Abbreviations

ASP aspect MI mimetic

F factitive NUC nuclear
IDEO ideophone P particle
IF illocutionary force PP pre-/post-positional phrase
L linker RP reference phrase
LSC layered structure of clause TNS tense
9.1 Introduction
The syntax of adverbs has triggered much controversy both within and
across theories (see, for example, Alexiadou 1997; Cinque 1999; Engels
2012; Ernst 2002, 2020).* The central concern includes the ability to offer
a principled account of (i) which positions an adverb can occupy within a
syntactic representation and (ii) what determines the ordering of mul-
tiple adverbs in sentence structure. Reflecting on recent developments in
the generative approach to adverbial syntax, Ernst notes the critical
consensus that ‘the semantics of individual adverbs is an important
determinant – perhaps the main determinant of their ordering’ (2014:
108). While Ernst is considering generative syntax, the essence of his
* I am grateful to Robert Van Valin, Jr. and to a reviewer for their valuable comments; they greatly improved the quality of
the paper. I thank James Watters for clarifying points on the Tepehua ideophones, and Janis Nuckolls for confirming the
point on the Quechua data. Thanks also go to Elizabeth Thompson for her editorial suggestions. The remaining
shortcomings are, of course, solely my responsibility.

404 KIYOKO TORATANI
reflection has close affinity with how Role and Reference Grammar (RRG)
analyses adverbs: it too considers semantics a critical component in
theorizing the syntax of adverbs. This affinity may not be accidental, as
a series of Ernst’s own work (e.g. 2002), characterized by him as a ‘scopal
approach’ (Ernst 2014), is rooted in Jackendoff (1972), and RRG draws on
insights from that same work.
Assuming that adverbs can be defined as ‘modifiers of constituents other
than nouns’ (Schachter 1985: 20), the first part of this chapter offers a
foundational sketch of the RRG approach to adverbs using English data,
drawing on Van Valin (2005) and Van Valin and LaPolla (1997), with a focus
on their subsets, including aspectual adverb (completely), manner adverb
(clumsily), temporal adverb (yesterday), evidential adverb (evidently), and
other clausal adverbs. It is pointed out that adverbs modify distinct layers
of the clause – the nucleus, the core, and the clause – and when the
sentence contains multiple adverbs, their appearance corresponds to the
order of the layers they modify in terms of the distance from the verb. The
second part of the chapter moves to an exemplification of how the RRG
approach can be used to analyse the ideophonic adverb, which is typically
a type of manner adverb (Schachter 1985: 21).1 While the category status of
ideophones remains disputed,2 it has frequently been noted that some
function as adverbs, appearing in the same syntactic position as adverbs
of prosaic words (Beck 2008: 41; Bobuafor 2013: 351; Ibarretxe-Antuñano
2006: 19; Nuckolls 1996: 72, among others). For instance, observing an
example like (1), wherein the ideophone expresses a manner of running,
Schaefer notes: ‘Emai IA [i.e. ideophonic adverb] forms assume the canon-
ical syntactic position of adverbs. They occupy [the] postverbal position’
(2001: 341).
(1) Emai (Nigeria’s Edoid group; Schaefer 2001: 347)

óli ómohe láí nyényényé
the man run-f ideo (with a dash)
‘The man dashed off’/‘the man ran off with a dash’
Despite the pervasive presence of ideophones in the languages of the

world (Voeltz and Kilian-Hatz 2001), virtually no work has analysed ideo-
phonic adverbs from a cross-linguistic perspective applying a specific syn-
tactic theory to the data. This chapter represents the first step toward
closing the gap by presenting an initial RRG view of the syntax of
ideophonic adverbs.
The chapter is organized as follows. Section 9.2 lays out theoretical
assumptions in RRG’s treatment of adverbs (Van Valin 2005; Van Valin and
LaPolla 1997), explaining the syntactic behaviour and properties of adverbs.
This is followed by a brief precis of the semantic representation of adverbs.
Section 9.3 turns to characteristics of ideophonic adverbs. Section 9.4 con-
tains concluding remarks.

Adverbs, Mimetics and Ideophones 405
9.2 Theoretical Assumptions
9.2.1 Structural Representation of Clause Structure

9.2.1.1 Layered Structure of the Clause (LSC)
RRG’s approach to adverbs is grounded in the conception of the layered
structure of the clause (LSC). Figure 9.1 previews an example of the repre-
sentation of the sentence, Leslie has evidently been slowly immersing herself
completely in the new language.
Figure 9.1 consists of two parts. The top part of the representation,
called the ‘constituent projection’ of the LSC, is centred on three primary
syntactic units: the clause (the outermost layer), the core (the mid layer),
and the nucleus (the innermost layer). The clause contains the core,
which, in turn, contains the nucleus and the arguments of the predicate,
categorized as the ‘reference phrases’ (Van Valin 2008). The nucleus
contains the predicate, usually a verb or an adjective. The peripheries
contain a modifier phrase (MP) (Van Valin 2008: 172) whose nucleus is
usually an adverb or an adjective. Each layer of the clause can be modi-
fied by a periphery, as indicated by an arrow in the figure. In Figure 9.1,
SENTENCE
PERIPHERY CLAUSE
PERIPHERY CORE
RP MP MP NUC PERIPHERY PP
COREP
COREM COREM PRED RP MP
NUCP RP
NUCM NUCM COREM CORER
PRED
ADV ADV V NUCM PERIPHERYR NUCR
ADV P MP N
Leslie has evidently been slowly immersing herself completely in the new language.
V N
NUC ASP NUCR
ASP NUC CORER

CORE DEF RP
TNS CLAUSE
IF CLAUSE
SENTENCE
Figure 9.1 Example of LSC (modified from Figure 1.13, Van Valin 2005: 22)

406 KIYOKO TORATANI
evidently modifies the clause, slowly modifies the core, and completely modi-
fies the nucleus. A brief note on the term ‘periphery’ is in order, as it is
sometimes confused with a concept in prototype theory (e.g. Lakoff 1987),
whereby membership in a category is characterized as ‘central’ (i.e. proto-
type) vs. ‘peripheral’ (non-prototypical), or given a configurational region
of ‘central’ (i.e. core) vs. ‘peripheral’ (i.e. margin). As Figure 9.1 makes
clear, in RRG, ‘periphery’ means neither a ‘non-prototypical member’ nor
a ‘margin’ but refers to a structural unit containing a modifier such as an
adverb (Van Valin 2005: 4).
The bottom part of the representation in Figure 9.1 is called the ‘operator
projection’. Operators refer to ‘grammatical categories . . . [that] modify the
clause and its parts’ (Van Valin and LaPolla 1997: 40). They are posited to
have scope over a specific layer of the clause:
The nuclear operators have scope over the nucleus; they modify the
action, event or state itself without reference to the participants. Core
operators modify the relation between a core argument, normally the
actor, and the action . . . Clausal operators, as the name implies, modify
the clause as a whole.
(Van Valin 2005: 8–9)
Again, the arrows in the figure indicate the modificational relations. Inci-
dentally, Figure 9.1 contains only three grammatical categories (ASP, TNS, IF
(i.e. aspect, tense, illocutionary force)), but RRG posits nine operator categor-
ies: (i) aspect, (ii) negation, (iii) directional, (iv) event quantification,
(v) modality, (vi) status,3 (vii) tense, (viii) evidentials, and (ix) illocutionary
force (Van Valin 2005: 12). Each modifies a designated layer (or layers) of the
clause, as represented in Figure 9.2: for instance, aspect is a nuclear oper-
ator. Some of the operator categories become relevant when we discuss the
syntactic characteristics of ideophones.
NUC Aspect
Negation
Directionals
CORE Directionals
Event quantification
Modality
Negation
CLAUSE Status
Tense
Evidentials
Illocutionary Force
SENTENCE
Figure 9.2 Operator projection (adapted from Van Valin 2005: 12)

9.2.1.2 Adverb Ordering

Although adverbs are not operators (Van Valin 2005: 19),4 like operators, they
are assumed to modify a specific layer of the clause, as exemplified in (2).
(2) Layer Examples of adverbs modifying the specific layer (English)

Nucleus: aspectual: completely, continuously
Core: pace: quickly; manner: gently; temporal: yesterday
Clause: epistemic: probably; evidential: evidently;
evaluative: unfortunately
The distribution of these adverbs is semantically motivated. Nuclear

adverbs modify the predicate in the nucleus. For instance, aspectual
adverbs such as completely are nuclear adverbs, because they add infor-
mation about the aspectual unfolding of the event denoted by the predi-
cate. Core adverbs modify the semantic content of the event denoted by
the predicate and the core argument(s). For instance, ‘subject-oriented’
adverbs such as clumsily are core adverbs, as they add information on how
the event participant performs the action. Lastly, clause adverbs modify
the entire proposition expressed by the clause. Evaluative adverbs such as
unfortunately, a type of ‘speaker-oriented’ adverb, are clausal, as they are
concerned with the speaker’s evaluation of the propositional content.5
A frequently noted aspect of adverbs in English is that they are not free to
occur anywhere within the sentence to yield a given reading. Drawing on
Jackendoff (1972) and McConnell-Ginet (1982), Van Valin (2005: 20) discusses
the case of manner adverbs, which ‘interact in an important way with the
tense operator’: that is, ‘those [adverbs] which occur before the tense oper-
ator can be construed as clausal modifiers, while those occurring after tense
cannot be’ (Van Valin 2005: 20). This is illustrated in (3).
(3) a. Ruth hid the cash cleverly.

b. Cleverly, Ruth hid the cash.
c. The manner in which Ruth hid the cash was clever. (¼3a)
d. The fact that Ruth hid the cash was clever. (¼3b)
e. Ruth cleverly hid the cash.
(3a) has the adverb cleverly after tense. It can have the manner reading given in
(3c) but lacks the reading of a clausal modifier given in (3d). By contrast, (3b)
has the adverb before tense, in which case, the reading as a clausal modifier
(3d) is available but the manner reading (3c) is absent. Meanwhile, (3e) locates
the adverb immediately before the verb, generating both readings.
If a sentence contains multiple adverbs of different types, their distribu-
tion is constrained by the ordering of the layers to which they are related.
That is, ‘adverbs related to more outer operators occur outside of adverbs
related to more inner operators’ (Van Valin 2005: 20). In the case of English,
with its ability to place adverbs either preverbally or postverbally, the order
of adverbs is affected by their position relative to the verb (cf. Ernst 2002:
154, 2014: 114–115).

408 KIYOKO TORATANI
First, consider the case of the core adverb occurring preverbally. (4) shows
the predicted order of three types of adverbs, and (5) shows different pat-
terns of a sentence containing the three adverbs in different positions.
(4) Order of adverbs when the core adverb precedes the verb:
clausal > core > nuclear
e.g. evidently [evidential: clausal] > slowly [pace: core] > completely
[aspectual: nuclear]
(5) a. Evidently, Leslie has slowly been completely immersing herself in

the new language.
b. Leslie has evidently been slowly immersing herself completely in
the new language.
c. * Evidently, Leslie has completely been slowly immersing herself in
the new language.
d. * Slowly, Leslie has evidently been completely immersing herself in
the new language.
e. * Slowly, Leslie has completely been evidently immersing herself in
the new language.
f. * Completely, Leslie has evidently been slowly immersing herself in
the new language.
g. * Completely, Leslie has slowly been evidently immersing herself in
the new language.
Van Valin (2005: 20)
In (5a), the three adverbs appear preverbally, whereas in (5b), the clausal
adverb evidently and the core adverb slowly occur preverbally, but the nuclear
adverb completely is postverbal. The adverbs in (5a)–(5b) observe the predicted
order, rendering the sentences grammatical. In contrast, (5c)–(5g) are judged
ungrammatical, as the adverbs disobey the predicted order given in (4): for
example, in (5c), when the nuclear adverb completely precedes the core
adverb slowly, the sentence is infelicitous; in (5d), when the core adverb
slowly precedes the clausal adverb evidently, the sentence is similarly
infelicitous.
A similar situation arises when the core adverb occurs postverbally. (6)
shows the predicted order, and (7) gives examples with adverbs occurring in
different positions.
(6) Order of adverbs when the core adverb follows the verb:
nuclear < core < clausal
e.g. completely [aspectual: nuclear] < slowly [pace: core] < evidently
[evidential: clausal]
(7) a. Leslie has been immersing herself completely in the new language
slowly evidently.
b. Leslie has been completely immersing herself slowly in the new
language evidently.
c. * Leslie has been immersing herself slowly in the new language
completely evidently.

d. * Leslie has been immersing herself completely in the new language

evidently slowly.
e. * Leslie has been immersing herself evidently in the new language
completely slowly.
f. * Leslie has been immersing herself slowly in the new language
evidently completely.
g. * Leslie has been immersing herself evidently in the new language
slowly completely.
Van Valin (2005: 20–21)
Among the different patterns, (7a) and (7b) are felicitous, wherein the three
adverbs appear as nuclear, core, and clausal adverbs, in that order, toward the
end of the sentence (although the nuclear adverb appears before the verb in
(7b)). In contrast, the sentences in (7c)–(7g) are infelicitous; they do not observe
the order in (6). For instance, in (7c), the nuclear adverb completely occurs
between the core adverb slowly and the clausal adverb evidently, and in (7d)
the core adverb slowly appears after the clausal adverb evidently.
These examples show that the ordering constraints work in both direc-
tions, substantiating the argument that ‘the scope constraints require that
the nuclear adverb be closer to the verb than the core adverb, and likewise
for the core adverb with respect to the clausal adverb’ (Van Valin 2005: 21).
These relationships are readily depicted in the LSC. Figure 9.3 shows the
SENTENCE
CLAUSE PERIPHERY
CORE PERIPHERY
MP
RP NUC PERIPHERY PP MP
RP
COREP COREM
MP COREM
NUCP RP
PRED NUCM
COREM CORER NUCM
PRED
PERIPHERYR NUCR
V NUCM ADV
ADV
ADV P MP N
Leslie has been immersing herself completely in the new language slowly evidently.
V N
NUC ASP NUCR
ASP NUC CORER
CORE
DEF RP
TNS CLAUSE
IF CLAUSE
SENTENCE
Figure 9.3 Structure of (7a) (modified from Figure 1.14, Van Valin 2005: 22)

410 KIYOKO TORATANI
constituent and operator projection of (7a) (see Figure 9.1 for the syntactic
representation of (5b)).
Although the examples discussed to this point are English, the fundamen-
tal point about adverb order is expected to be cross-linguistically valid. That
is, regardless of which language is under investigation, the relative order of
adverbs should observe the order of the layers they modify: for example,
adverbs modifying the entire clause should appear away from the predicate,
and adverbs modifying the predicate should appear in closer proximity to
the predicate.
It is worth pointing out that temporal adverbs such as tomorrow (in their
default position) are core modifiers, not clausal modifiers, although the
meaning of temporal adverbs is related to tense, and tense is a clausal
operator (Van Valin 2005: 19; Van Valin and LaPolla 1997: 32, 426–428). This
point is elaborated by Bohnemeyer and Van Valin (2017), who note that
‘verbal cores are inherently constituents that describe (sub)events and that
each core . . . has its own time-positional modifier’ (2017: 159). For instance,
in (8), on Monday modifies the event of persuading, whereas on Friday modi-
fies the event of visiting.
(8) Tom persuaded Sally on Monday to visit her sister on Friday.

(Bohnemeyer and Van Valin 2017: 159)
It should be emphasized that the occurrence of temporal adverbs in the

periphery is a default case; they can also occur in the ‘pre-detached
position’, which is ‘outside of the clause but within the sentence’ (Van Valin
and LaPolla 1997: 36). For example, consider (9).
(9) a. Sam decided to leave tomorrow.

b. Yesterday, John did not show the book to Mary.
c. *Tomorrow, Sam decided to leave.
(Van Valin and LaPolla 1997: 170)
In (9a), tomorrow appears in the default position, whereas in (9b), yesterday

appears in the pre-detached position. When a temporal adverb appears in
the latter position, it is no longer contained in the periphery and no longer
modifies the core: ‘If a temporal adverb occurs in the left [i.e. pre]-detached
position, then it is a clausal modifier’ (Van Valin and LaPolla 1997: 170). This
point can be illustrated by comparing (9a) and (9c). If a temporal adverb
always remains a core modifier regardless of its position within the sen-
tence, tomorrow in (9a) should be able to be fronted, yielding the meaning of
(9a). However, this is not the case. As indicated by the unacceptability of (9c),
if the temporal adverb tomorrow appears in the pre-detached position, the
ensuing clausal reading implies that the event of deciding will take place
tomorrow. However, this turns out to be incompatible with the information
of tense, which is past.
Now that we have noted the modificational unit of temporal adverbs, it
may be worth asking about possible constraints when multiple adverbs

modify the same layer of the clause. In fact, when discussing examples like
(10), Van Valin (p.c.) suggests semantic-based constraints may be at work;
more specifically, if three kinds of core adverbs co-occur (as in (10): a pace
adverb quickly, a manner adverb carefully, and a temporal adverb yesterday),
ordering constraints appear.
(10) a. John carefully opened the box quickly yesterday.

b. ?? John carefully opened the box yesterday quickly.
c. John quickly opened the box carefully yesterday.
d. ?? John quickly opened the box yesterday carefully.
e. * John yesterday opened the box carefully/quickly.
f. John opened the box carefully and quickly yesterday.
g. * John opened the box carefully and yesterday.
h. * John opened the box quickly and yesterday.
j. Yesterday John carefully opened the box quickly.
First, with respect to position, the pace adverb and the manner adverb seem
to be used interchangeably ((10a) vs. (10c)). Second, the pace adverb and the
manner adverb can be conjoined but neither can be conjoined with the
temporal adverb ((10f) vs. (10g/10h)). Third, in the postverbal position, the
temporal adverb should follow the manner adverb ((10c) vs. (10d)), and
preverbally, the temporal adverb should precede the manner adverb (10j).
The differences in acceptability in (10) strongly suggest some semantic
layering, at least within the core; however, a closer investigation is called
for, with more data.
9.2.2 Semantic Representation of Adverbs

Adverbs are represented by logical structures of ‘one-place predicates
which take a logical structure or subpart of a logical structure as their
argument, following the approach by Jackendoff (1972) and others’ (Van
Valin and LaPolla 1997: 162; cf. Van Valin 2005: 49). Which part of the
logical structure an adverb takes as its argument depends on the meaning
of the adverb, given its syntactic position within the sentence. A few
examples are given in (11).
(11) a. Yesterday, Chris ran to the park.

a′. yesterday′ (do′ (Chris, [run′ (Chris)]) & INGR be-at′ (park, Chris))
b. Pat elegantly closed the door slowly.

b′. [elegant′ (do′ (Pat, Ø))] CAUSE [slow′ (BECOME closed′ (door))]
c. The ice melted completely/The ice completely melted.

c′. BECOME complete′ (melted′ (ice))
Temporal adverbs like yesterday in (11a) take the entire logical structure as
their argument (Van Valin 2005: 49). Manner adverbs like elegantly in (11b)
take the activity component as their argument (indicating the elegant

412 KIYOKO TORATANI
manner in which Pat executed the action of closing), whereas the pace
adverb slowly in (11b) takes the accomplishment component as its argument
(indicating the slowness of the door’s closing event). Finally, in (11c), the
aspectual adverb completely takes the basic state component melted′ (x) as
its argument.
It is left for future work to evaluate whether this uniform treatment of
adverbs, that is, positing a one-place predicate as their semantic representa-
tion, can fully account for the characteristics of adverbs across the board,
either within a given language or cross-linguistically.
9.3 Syntax of Ideophonic Adverbs
As noted in Section 9.1, some ideophonic adverbs can express manner

(Schachter 1985: 21). This leads to the question of whether they pattern
in the same fashion as manner adverbs like clumsily: that is, whether
ideophonic adverbs are also core or clausal modifiers, as discussed in
Section 9.2. In answering this question, we first provide preliminary infor-
mation about Japanese ideophones, called mimetics, as the ensuing discus-
sion largely relies on Japanese data (Section 9.3.1). Second, we use
syntactically relevant semantic criteria to classify ideophones into core
and nuclear adverbs (Section 9.3.2). Third, we examine whether the classi-
fication holds if the ideophone appears in a given syntactic environment
(Section 9.3.3). Finally, we turn to a case where ideophones are not con-
tained in the periphery, drawing on data from a Totonac-Tepehua language
(Watters 2013) (Section 9.3.4).
9.3.1 Japanese Mimetic Adverbs

Japanese mimetic adverbs can be classed into three major forms, which we
will call ‘singleton’, ‘reduplicated’, and ‘ri-suffixed’, exemplified in (12).
(12) singleton: kon ‘a knock’

botat ‘a drip’
een ‘a short crying voice’
reduplicated: kurukuru ‘spinning’
batabata ‘continuously falling’
betabeta ‘sticky’
ri-suffixed: sukkari ‘entirely’
yukkuri ‘slowly’
pittari ‘tightly’
Aspect is argued to be part of the lexical meaning of the first two forms
(Akita 2009; Toratani 2005). Singletons typically express an event/state
lasting a short time, such as a semelfactive event, that is, a single-staged
event with no outcome (Smith 1997: 29) (e.g. kon ‘sound of a knock’), an

event with a culmination point (e.g. bokit ‘a snap; a manner of a sudden

breakage of an object’), or an atelic event lasting a short time (e.g. een ‘a
short-lasting crying voice’). They are repeatable: for example, botat ‘a drip (of
liquid)’ can be repeated, as in botat botat ‘drip, drip’, or botat botat botat ‘drip,
drip, drip’ and so forth, with a phonological break in between.
Reduplicated forms such as kurukuru ‘manner of continuous spinning’
have a fully reduplicated structure with no word-internal phonological
break. They can express the iteration of a cyclic (e.g. kurukuru ‘iteration of
a spinning of an object’) or telic event (e.g. batabata ‘continuously falling’), or
the continuation of a dynamic (e.g. ziroziro ‘continuation of one’s staring
event’) or static condition (e.g. betabeta ‘continuation of a sticky bodily
sensation’).
Ri-suffixed mimetics (ending in the form -ri) typically express a non-event-
specific condition, such as sikkari ‘firmly’ or yukkuri ‘slowly’, and are ‘de-
ideophonized’ (cf. Dingemanse 2017), in the sense that they lack the expres-
sivity and the aspectual character present in many singletons and
reduplicated forms.
When mimetic adverbs occur in a sentence, some forms are marked by to
‘quotative (elsewhere)’ (to-marked) or not marked by to (Ø-marked),
depending on the morphological shape of the mimetics, as exemplified
in (13).
(13) a. Zyaguti-kara sizuku-ga potot-to (*Ø) oti-ta.

faucet-from drop-nom mimetic-p drip-pst
‘A drop (of water) dripped from the faucet potot.’
b. Zyaguti-kara sizuku-ga potot potot-to (*Ø) oti-ta.
‘(Two) drops (of water) dripped from the faucet potot potot.’
c. Zyaguti-kara sizuku-ga potopoto-to/Ø oti-ta.
‘Drops (of water) dripped continuously from the faucet.’
d. Zyaguti-kara sizuku-ga yukkuri-to oti-ta.
‘Drops (of water) dripped slowly from the faucet.’
On the one hand, singletons are obligatorily marked by to, regardless of how
many times the form is instantiated: once (13a), twice (13b), and so forth. On
the other hand, reduplicated forms (13c) and ri-suffixed forms (13d) can be
marked by to or Ø, yielding no truth-conditional differences between the
alternative markings (Hamano 1998, among others; see Akita and Usuki
2016 for a discussion of ‘optionality’ of to-marking). As (13) suggests, the
most frequently occupied position of mimetics of any morphological shape
is immediately preverbal (cf. Toratani 2017: 36–41). However, because Japan-
ese, an OV language, has a flexible word order, clause-internal phrases,
including mimetics, can scramble rather freely as long as the verb comes
at the end of the clause.

414 KIYOKO TORATANI
9.3.2 Syntactically Relevant Semantic Criteria

Certain grammatical categories and the concept of hyponymy provide a
semantic diagnostic for determining which layer of the clause the ideo-
phone potentially modifies.
9.3.2.1 Grammatical Categories

As Section 9.2 notes, operators modify a specific layer of the clause (see
Figure 9.2). If an ideophone expresses the same concept as an operator, it
is reasonable to posit that it modifies the same layer of the clause. The
question is whether an ideophone can express any concept equivalent to
that expressed by an operator. The answer is yes, at least in the following
two grammatical categories.
The first category is event quantification, which expresses multiple actions
of the verb and is classed as a core operator. As (14) shows, some Japanese
mimetics express the concept of event quantification.
(14) a. Hito-ga heya-ni hait-te-ki-ta.

person-nom room-dat enter-l-come-pst
‘A person came into the room.’
b. Hito-ga zorozoro heya-ni hait-te-ki-ta.
person-nom mimetic room-dat enter-l-come-pst
‘People came into the room in line one after another.’
(14a) represents the base sentence with no mimetic. The noun hito ‘person/
people’ has an ambiguous reading, in that it can be singular or plural, as nouns
are not obligatorily marked for number in Japanese; yet the default interpret-
ation of the sentence is a single event of coming in. In contrast, (14b)’s mimetic
zorozoro implies multiple events of coming in, thereby depicting the motion of
a crowd of people lining up and moving forward. Therefore, this mimetic is
analysed as a core modifier, as it expresses event quantification.
Other languages have ideophones lexically encoded with the notion of
mass, as exemplified below.
ˇ
(15) a. Chichewa (Bantu; Kulemeka 1993: 224)
u:nji: ‘gather in a mass’
b. Luwo (Western Nilotic, South Sudan; Storch 2014: 44)
m ̀ rm ̀ r
c c ‘a lot of people/cattle moving together’
c. Siwu (Ghana-Togo Mountain; Dingemanse 2011: 48)
ɣèèè ‘animals swarming in great numbers’
The sentence from Pastaza Quechua in (16) makes an analogous point. It

contains dzhawww, a variant of dzawn, meaning an ‘action, process, or event
that involves a clustering together of individual agents, such as people,
birds, bats, or insects’ (Nuckolls 1996: 148).
(16) Pastaza Quechua (Ecuador; Nuckolls 1996: 149)

Putan dzhawww hatari-ra!
fly ideo rise up-pst
‘The flies rose up dzhawww!’

According to Nuckolls, ‘dzawn describes the way flies that had gathered on a
molting snake rose up in a swarm when a person approached’ (1996: 149).
The fact that putan ‘fly’, glossed in the singular, is interpreted as plural in
(16) suggests the ideophone plays a role in assigning the mass interpretation.
If this is the case, the ideophone in (16) is a core modifier.
The second grammatical concept applicable to ideophones is aspect. As
noted in (2), aspectual adverbs are assumed to be nuclear modifiers (Van
Valin 2005: 49). Examples are continuously and completely. The assumption is
that ‘aspectual adverbs modify the basic state or activity predicate’ (Van
Valin 2005: 49): e.g. melt completely: BECOME complete′ (melted′ (x)). If ideo-
phonic adverbs can express a concept similar to completely and continuously
and modify the basic state or activity predicate, it seems reasonable to posit
them as nuclear modifiers. Some languages seem to have them, or at least
the English glosses suggest this possibility (e.g. wic ‘eat all up’ (Luwo, Storch
ˇ
2014: 47), psí:tí: ‘completely finished’ (Chichewa, Kulemeka 1993: 252), tdip
‘manner of covering completely’ (Didinga, a Southwest Surmic language,
Sudan, de Jong 2001: 133)). In Japanese, no mimetics precisely express the
meaning of completely and continuously. However, as discussed in Section
9.3.3.2, some mimetics affect the interpretation of the lexical aspect of the
verb when they occur adjacent to the verb and, as such, seem to qualify as
possible nuclear modifiers.
9.3.2.2 Hyponymy
The second semantic criterion to determine which layer of the clause ideo-
phones modify is hyponymy. It has long been noted that some Japanese
mimetics co-occur with a limited set of verbs (Hirose 1981). For instance,
tekuteku ‘plodding’ or yotiyoti ‘toddling’ typically co-occur with aruku- ‘walk’.
In such cases, the mimetic–verb relationship can be characterized in terms of
hyponymy (Toratani 2007: 325–327), where the mimetic is considered the
hyponym and the verb is the hyperonym of the mimetic, as tekuteku ‘plodding’
and yotiyoti ‘toddling’ both express a kind of a walking event.6
In other languages as well, some ideophones collocate or co-occur with a
limited number of verbs/adjectives (Childs 1994: 188; Creissels 2001: 78). For
instance, in Didinga (a Southwest Surmic language in Sudan), the ideophone
ɪðaatʃ ɪðaatʃ ‘manner of swallowing easily’ co-occurs with kú ‘swallow’ (de
Jong 2001: 136), while in Emai (Nigeria’s Edoid group), the ideophone
ghóighói ‘glistering-ly’ co-occurs with jín ‘shine’ (Schaefer 2001: 351), with
the former apparently the hyponym of the latter in each pair.
The hyponymy relationship highlights various ways in which ideophones
and core adverbs modify the predicate’s event. Regular core adverbs can co-
occur with a wide variety of predicates (clumsily walk/dance/write/sit etc.), as
they refer to a general property common to a group of predicates, such as
dynamism, conveying how the event participant performs the action with-
out changing the meaning of the action itself; meanwhile, when ideophones
co-occur with a limited set of predicates, they change the reading of the

416 KIYOKO TORATANI
action from a more or less neutral manner (e.g. ‘walking’) to a much more
detailed one (e.g. ‘toddling’). To interpret this difference in terms of the layer
of the clause the adverbs modify, we could say the semantic function of
ideophones closely parallels that of nuclear adverbs, in that their functions
are in line with the nuclear operators’ function of ‘modify[ing] the action,
event or state itself without reference to the participants’ (Van Valin 2005:
8–9). This parallelism suggests ideophones that co-occur with the verbs/
adjectives of their hyperonymous category are nuclear adverbs, unless they
refer to multiple event participants, in which case, the ideophones are core
modifiers.7
The preceding discussion may give the impression that all ideophones are
hyponyms (cf. Watson 2001: 393). This is not the case. To draw again on
Japanese data, while some mimetics are clearly hyponyms (candidates for
nuclear adverbs) as exemplified in (17), others are non-hyponyms (candidates
for core adverbs).8
(17) Hyponyms (candidates for nuclear adverbs)

a. yotiyoti ‘toddle-toddle’
b. musyamusya ‘munch-munch’
c. turuturu ‘slippery-slippery’
d. guruguru ‘spinning round and round’
(18) Non-hyponyms (candidates for core adverbs)

a. yukkuri ‘slowly’
b. sot ‘gently’
c. syonbori ‘dispiritedly’
d. bonyari ‘absentmindedly’
Non-hyponyms in (18) include a pace adverb, yukkuri ‘slowly’, and several

‘subject-oriented’ adverbs. The latter category includes general manner
adverbs such as sot ‘gently’ and a subset of ‘psychomimes’ (Martin 1975:
1025) (those expressing a psychological state) such as bonyari ‘absentmind-
edly’ (18d). An entailment test can be used to distinguish hyponyms from
non-hyponyms. If the utterance of an ideophone entails the occurrence of a
specific set of events, it can be judged a hyponym; otherwise, it is a non-
hyponym. For instance, the utterance of the hyponymous mimetic guruguru
‘spin-spin’ entails that something continuously rotates and is readily associ-
ated with a verb such as mawaru ‘turn’, but the utterance of the non-
hyponymous mimetic yukkuri ‘slowly’ does not entail a specific event
and thus disallows the instantaneous association available with the
hyponymous mimetic.
Other languages seem to have non-hyponymous types of ideophones as
well. For instance, Patent (1998: 196–197) discusses the case of Lai Chin (Sino-
Tibetan), noting that while typical ideophones have elaborate meanings,
some have very general meanings and can co-occur with a number
of different verbs, such as tsiam-maam ‘with effort/forcefully’ and leŋ-
maŋ ‘always’.

To sum up, we make four observations, expressed in (19), albeit with a

proviso: as all observations are based on meaning, they may not apply if an
overriding syntactic factor comes into play.
(19) Observation 1: An ideophone that expresses event quantification is a core

adverb.
Observation 2: An ideophone that affects the aspectual interpretation of
the basic state or activity predicate is a nuclear adverb.
Observation 3: A hyponymous ideophone that co-occurs with a clause-mate
verb/adjective from its hyperonym category without making
reference to a plural number of event participants is a
nuclear adverb.
Observation 4: A non-hyponymous ideophone is a core adverb.
These observations imply that ideophones are not clausal modifiers. This
possibility can be confirmed with Japanese data. As noted in Section 9.2.1.2,
typical clausal adverbs express the speaker’s evaluation or judgement of a
propositional content (e.g. unfortunately, probably). Mimetics do not seem to
convey these concepts. Furthermore, like mimetics, the large majority of ideo-
phones express sound, manner, and the state of an entity, detailing the actions
and states expressed by the clause-mate verbs/adjectives, but they do not seem
to convey any concepts paralleling those conveyed by clausal adverbs. This
suggests that ideophones are not clausal adverbs cross-linguistically as far as
the meaning is concerned, but this suggestion requires validation.9
9.3.3 Syntactic Positions of Ideophonic Adverbs

As the observations in (19) are based on meaning, the next question is
whether a syntactic unit modified by an ideophone can be maintained when
it actually appears in a sentence. To answer this question, we begin by
considering whether core adverbs can occur as clausal adverbs, just as
English manner adverbs such as clumsily can be core or clausal modifiers
(Section 9.3.3.1). Next, we examine whether nuclear adverbs can occur as
core adverbs (Section 9.3.3.2).
9.3.3.1 Possibility of a Core Adverb Occurring as a Clausal Adverb

The first question is whether ideophonic adverbs expressing manner can
yield an alternative evaluative (clausal) reading when they change position
within a sentence, just like English manner adverbs such as clumsily. (20)
shows this type of alternation is unavailable to Japanese mimetics. It is
assumed that the mimetic batan ‘a bang’ is a core adverb, as it is not a
hyponym of sime- ‘close’. Rather, it refers to the sound of the object (door), re-
enacting the event of the door’s having been shut.
(20) a. Taroo-ga to-o batan-to sime-ta.

Taro-nom door-acc mimetic-p close-pst
‘Taro closed the door with a bang.’

418 KIYOKO TORATANI
b. Taroo-ga batan-to to-o sime-ta.

Taro-nom mimetic-p door-acc close-pst
‘Taro closed the door with a bang.’
c. Batan-to Taroo-ga to-o sime-ta.
mimetic-p Taro-nom door-acc close-pst
‘With a bang Taro closed the door.’
(adapted from Toratani 2007: 11–12)
Example (20) contains the mimetic batan ‘(sound of ) a bang’ in three differ-
ent preverbal positions. As the identical gloss (with a bang) indicates, it yields
only one reading, that of manner, irrespective of where the mimetic occurs
within the sentence.
This unambiguity of reading is supported by the paraphrasability of the
meaning of the mimetic into a manner reading (21a) but not into an
evaluative (i.e. clausal adverb) reading (21b).
(21) a. Taroo-no to-no sime-kata-wa batan-to-dat-ta.

Taro-gen door-gen close-way-top mimetic-p-cop-pst
‘The way Taro closed the door was with a bang.’
b. *Taroo-ga to-o sime-ta-no-wa batan-to-dat-ta.

Taro-nom door-acc closed-pst-nmlz-top mimetic-p-cop-pst
‘The fact that Taro closed the door was with a bang.’
(adapted from Toratani 2007: 12)
In other words, the mimetic remains a core modifier and cannot be clausal,
even if it changes syntactic position. This leads to the following two predic-
tions. First, the mimetic should be able to co-occur with a clausal adverb:
more specifically, because the mimetic and the clausal adverb modify dis-
tinct layers of the clause, they will not cause a semantic clash, unlike two
clausal adverbs as in evidently and probably in *Evidently, John probably left
(Jackendoff 1972: 87). Second, the mimetic should observe the order con-
straint with respect to the clausal adverb. That is, the clausal adverb should
occur first, followed by the mimetic. (22) and (23) illustrate these points (the
mimetic gatyan ‘sound of a clank’ is a core modifier as it is not a hyponym of
otosu ‘drop’).
(22) a. Bukiyooni-mo kabin-o gatyan-to otosi-te.simat-ta.

clumsy-p vase-acc mimetic-p drop-asp-pst
‘Clumsily (he) dropped the vase with a clank.’
b. Bukiyooni-mo gatyan-to kabin-o otosi-te.simat-ta.
clumsy-p mimetic -p vase-acc drop-asp-pst
‘Clumsily (he) dropped the vase with a clank.’
(23) ?? Gatyan-to bukiyooni-mo kabin-o otosi-te.simat-ta.

mimetic-p clumsy-p vase-acc drop-asp-pst
‘With a clank clumsily (he) dropped the vase.’
(adapted from Toratani 2007: 13–14)

The first point is illustrated in (22) by the ability of the mimetic gatyan
‘sound of a clank’ to co-occur with the clausal adverb bukiyooni-mo ‘clum-
sily’.10 The second point (word order) is also illustrated in (22) by the
expected word order: the clausal adverb bukiyooni-mo ‘clumsily’ precedes
the mimetic (core/nuclear adverb). This is further supported in (23) by the
mimetic’s inability to cross the clausal adverb.
To reiterate, unlike English manner adverbs like clumsily, mimetic adverbs
do not yield a clausal reading, even if they change syntactic position.
Though verification is necessary, this seems to apply to ideophones
across languages.
9.3.3.2 Possibility of a Nuclear Adverb Occurring as a Core Adverb

Next, we examine whether ideophones that are semantically determined as
nuclear adverbs (those expressing aspect and/or all hyponymous ideophones
co-occurring with their clause-mate predicates from their hyperonymous
categories) are always nuclear adverbs, regardless of their position within
the sentence. To explore this question, we use Japanese data, drawing on
Tsujimura and Deguchi’s (2007) observation that the acceptability of the
sentence with a for/in-phrase (24a) is affected if a mimetic is added (24b).
(24) a. Mizu-o gohunkan/gohun-de nonda.

water-acc for/in 5 minutes drank
‘I drank water for/in five minutes.’
b. Mizu-o gohunkan/*?gohun-de gokugoku nonda.
water-acc for/in 5 minutes mimetic drank
‘I drank water (repeatedly) for/*?in five minutes.’
(Tsujimura and Deguchi 2007: 345)
Example (24a) is a simple sentence containing a transitive verb nomu ‘drink’

and a for/in-phrase, diagnostics for atelicity and telicity, respectively. The
verb can co-occur with either phrase, making it either atelic or telic. In
contrast, in (24b), the verb must be interpreted as atelic when the mimetic
gokugoku ‘gulp-gulp’ is present, as indicated by the acceptability of the for-
phrase but the unacceptability of the in-phrase. Tsujimura and Deguchi note
that ‘the sense of repetition associated with reduplicated mimetics . . .
affects the telicity of the sentences in which they occur’ (Tsujimura and
Deguchi 2007: 344). In RRG terms, this sensitivity of the mimetic to the
verbal aspect implies that the mimetic is a nuclear adverb.
This, however, does not mean mimetics always function as nuclear
adverbs. Observing the pattern in (24), discussed in Tsujimura and Deguchi
(2007: 344), Toratani (2007) points out that the mimetic compatible with the
atelicity reading of the verb appears in a particular environment. First, the
mimetic gokugoku is Ø-marked. Second, it occurs in the immediately pre-
verbal position. Third, the verb comes from the mimetic’s hyperonym
category. Of these, if the first two conditions are altered, the mimetic no
longer participates in specifying the verb’s aspect, as illustrated in (25), in

420 KIYOKO TORATANI
which the to-marked version of the mimetic moved out of the immediately
preverbal position.
(25) Kodomo-ga gokugoku-to mizu-o gohunkan/gohun-de non-da.

child-nom mimetic-p water-acc for/in 5 minutes drink-pst
‘The child, in a gulping manner, drank the water in/for five minutes.’
(Toratani 2007: 333)
The mimetic gokugoku in (25) is reduplicated, expressing aspectual continu-

ity. Furthermore, it co-occurs with a verb of its hyperonymous category, thus
satisfying the semantic criterion for nuclear adverbs. However, it does not
affect the aspectual reading of the verb; thus, satisfying the semantic criter-
ion alone (Observations 2 and 3 in (19)) is insufficient to claim that the
mimetic is a nuclear adverb. The mimetic must satisfy the morphosyntactic
condition: it must appear in the immediately preverbal position with no
elements intervening between it and the verb.
Although a more thorough investigation is necessary, those mimetics and
ideophones affecting the aspectual readings of their clause-mate predicates
can be posited to be nuclear adverbs; otherwise they are core adverbs. In
light of this point, the semantic-based observations in (19) can be revised as
(26), which is a working hypothesis for ideophones that co-occur with the
predicates of their hyperonymous categories (without making reference to a
plural number of event participants).
(26) The ideophone that specifies the aspect of the predicate is a nuclear adverb;
all other ideophones are core adverbs.
Ideophonic nuclear adverbs likely occur adjacent to the predicate belonging

to the hyperonymous category of the ideophone, but this requires confirm-
ation in individual languages.
Figure 9.4 represents a nuclear adverb, and Figure 9.5 shows a core adverb
(the internal layers of RPs, PPs and MPs are simplified; the operator projec-
tions are omitted).
SENTENCE
CLAUSE
PERIPHERY CORE
RP PP PERIPHERY NUC
MP PRED
MI-ADV V
Mizu-o gohunkan gokugoku non-da.
water-ACC for.5.min MIMETIC drink-PST
‘The child gulped water for five minutes.’
Figure 9.4 The mimetic as a nuclear adverb (adapted from Toratani 2007: 333)

SENTENCE
CLAUSE
PERIPHERY CORE
NUC
RP MP RP PP
PRED
MI-ADV V
Kodomo-ga gokugoku-to mizu-o gohunkan/gohun-de non-da.
child-NOM MIMETIC-P water-ACC for/in 5 minutes drink-PST
‘The child, in a gulping manner, drank the water in/for five minutes.’
Figure 9.5 The mimetic as a core adverb (modified from Toratani 2007: 334)
As Figures 9.4 and 9.5 show, when they are at clause-internal positions,
mimetics are contained in the periphery and modify a specific layer of
the clause.
9.3.3.3 Nuclear-Internal Modifier

The foregoing discussion suggests ideophonic adverbs modify the nucleus
and the core, unlike the clumsy-type manner adverbs, which can modify the
core and the clause. Recent RRG work by Watters (2013) suggests ideophones
can fall into another type, not found in the pattern of ideophonic adverbs
discussed so far.
Example (27) shows how ideophones are used in a Totonac-Tepehua lan-
guage. Noting that nothing can intervene between the ideophone and the
verb, Watters (2013) proposes a schematic structure (28) capturing the syn-
tactic tightness of ideophone and verb.
(27) Tlachichilco Tepehua (Totonacan; Watters 2013: 32)

a. sk’uli ʔu-y ki-makaː
itchily eat-ipfv 1poss-hand
‘it itches my hand’
b. spuy tsuku-y xkaːn
in.drops be-ipfv water
‘it’s sprinkling’
c. ʃtay ʔan-Ø
circularly go-ipfv
‘it rotates’
(28) [[X]ADV:IDEOPH [Y]VERB]NUC
Interestingly, in these examples, the ideophones co-occur with a very par-

ticular verb; unlike the case of Japanese mimetics, however, the verbs do not
belong to the hyperonymous category of ideophones, such as ‘scratch’ for
‘itchily’ and ‘turn’ for ‘circularly’. According to Watters (p.c.), the language
has ‘a much more limited verb vocabulary compared to English’ and must,
therefore, recruit semantically lighter verbs, such as ‘be’ and ‘go’, to partner

422 KIYOKO TORATANI
with ideophones, with the ideophones assuming the more substantial

semantic load.11 This implies reconceptualization is necessary to character-
ize the structural position of ideophones in (27) since, unlike the subsump-
tion relation noted for Japanese mimetics, the verb can no longer subsume
the semantics of the ideophone. Absence of the subsumption relation in (27)
indicates the ideophone becomes a critical component in the expression of a
given meaning. As Watters (p.c.) notes, the ideophones in (27) are, in theory,
adjuncts, in that the sentence remains grammatical even without them, but
the omission of the ideophones drastically changes the meaning, suggesting
they are required to portray the intended event type.
How should (28) be represented in terms of the LSC? There seem to be
three possibilities, as shown in Figure 9.6, using (27c) as an example.
First, Figure 9.6(c) models according to the representation of nuclear
adverbs of prosaic words, in which the adverbs are adjuncts. Since the
ideophone in (27c) is the required part of the sentence to convey the
intended meaning, this cannot be the correct representation. Next,
Figure 9.6(b) indicates that the ideophone and the verb are each dominated
by a PRED node of their own, thus implying each element expresses its own
meaning: that is, something ‘(moves) circularly’, and it ‘goes’. This cannot be
the correct representation either, because the ideophone and the verb are
semantically dependent on each other to express the intended meaning of
‘it rotates’. This semantic dependency of the ideophone and the verb
suggests they form a complex predicate, which, in turn, implies that
Figure 9.6(a) is the correct representation, with the ideophone and the verb
dominated by the same PRED node in the nucleus. The representation is also
consistent with the understanding that nothing can intervene between the
ideophone and the verb.
To sum up, the Tepehua ideophones in (27) lend themselves to a category
distinct from the ideophonic adverbs discussed in the previous subsections.
They enter into a particular semantic relationship with the verbs. The verbs
with which they co-occur are not as ‘light’ as DO/SAY verbs, often said to
(a) SENTENCE (b) SENTENCE (c) SENTENCE

CLAUSE CLAUSE CLAUSE
CORE CORE CORE
NUC NUC PERIPHERY NUC
PRED PRED PRED MP PRED
IDEO-ADV V IDEO-ADV V IDEO-ADV V
tay an- tay an- tay an-

circularly go-IPFV circularly go-IPFV circularly go-IPFV
‘it rotates’ ‘it rotates’ ‘it rotates’
Figure 9.6 Three possible representations for (27c)

be required by ideophones to function as a predicate (Childs 1994: 187)

(e.g. nikoniko-suru smilingly-do ‘smile’, an example from Japanese); nor are
they as ‘heavy’ as verbs in the ideophones’ hyperonym category, the pattern
commonly observed in Japanese mimetics (e.g. nikoniko warau [smilingly
laugh] ‘smile’). In other words, the Tepehua ideophone–verb semantic rela-
tionship constitutes an intermediary type, a category thus far undiscussed.
9.4 Concluding Remarks
This chapter outlines the RRG approach to adverbs, paying particular atten-
tion to ideophonic adverbs. It has introduced relevant RRG assumptions on
adverbs: (i) that adverbs contained in the periphery and represented in the
‘constituent projection’ of the layered structure of the clause may modify all
three layers of the clause, that is, the nucleus, the core, and the clause;
(ii) that the order of adverbs modifying a distinct layer of the clause observes
the order of the layers of the clause with respect to the position of the verb,
as the nuclear adverb is constrained to occur closer to the verb than the core
adverb, which, in turn, is constrained to occur closer to the verb than the
clausal adverb.
The RRG approach to adverbs remains rather preliminary (cf. Cortés-
Rodríguez and Rodríguez-Juárez 2019), dealing only with basic phenomena.
More analyses are certainly required, for instance, to identify the ordering
constraints among the adverbs modifying the same layer, considering their
motivations, or to substantiate how the logical structure of ideophonic
adverbs can be represented (for example, does this call for a one-place
predicate just like adverbs of prosaic words?) to capture the unique seman-
tics of ideophones.
References
Akita, Kimi. 2009. A Grammar of Sound-Symbolic Words in Japanese: Theoretical

Approaches to Iconic and Lexical Properties of Mimetics. Unpublished PhD
dissertation, Kobe University.
Akita, Kimi and Takeshi Usuki. 2016. A constructional account of the
‘optional’ quotative marking on Japanese mimetics. Journal of Linguistics
52: 245–275.
Alexiadou, Artemis. 1997. Adverb Placement: A Case Study in Antisymmetric
Syntax. Amsterdam: John Benjamins.
Beck, David. 2008. Ideophones, adverbs, and predicate qualification in Upper
Necaxa Totonac. International Journal of American Linguistics 74: 1–46.
Bobuafor, Mercy. 2013. A Grammar of Tafi. Unpublished PhD dissertation,
Leiden University.

424 KIYOKO TORATANI
Bohnemeyer, Jürgen and Robert D. Van Valin, Jr. 2017. The macro-event
property and the layered structure of the clause. Studies in Language 41
(1): 142–297.
Childs, Tucker G. 1994. African ideophones. In Leanne Hinton, Johanna
Nichols and John J. Ohala (eds.), Sound Symbolism, 178–204. Cambridge:
Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Per-
spective. Oxford: Oxford University Press.
Cortés-Rodríguez, Francisco, J. and Carolina Rodríguez-Juárez. 2019. The
syntactic parsing of ASD-STE100 adverbials in ARTEMIS. Revista De Lingüís-
tica y Lenguas Aplicadas 14: 59–79.
Creissels, Denis. 2001. Setsuwana ideophones as uninflected predicative
lexemes. In E. F. K. Voeltz and C. Kilian-Hatz (eds.), Ideophones, 76–85.
Croft, William and D. Alan Cruse. 2004. Cognitive Linguistics. Cambridge:
De Jong, Nicky. 2001. The ideophone in Didinga. In E. F. K. Voeltz and C.
Kilian-Hatz (eds.), Ideophones, 121–138. Amsterdam: John Benjamins.
Dingemanse, Mark. 2011. The Meaning and Use of Ideophones in Siwu. Unpub-
lished PhD dissertation, Radboud University Nijmegen.
Dingemanse, Mark. 2017. Expressiveness and system integration: On the
typology of ideophones, with special reference to Siwu, STUF – Language
Typology and Universals 70(2): 363–385.
Dingemanse, Mark. 2019. ‘Ideophone’ as a comparative concept. In K. Akita
and P. Pardeshi (eds.), Ideophones, Mimetics, and Expressives, 13–33. Amster-
Doke, Clement Martyn. 1935. Bantu Linguistic Terminology. London: Longmans,
Green, and Co.
Engels, Eva. 2012. Optimizing Adverb Positions. Amsterdam: John Benjamins.
Ernst, Thomas. 2002. The Syntax of Adjuncts. Cambridge: Cambridge University
Press.
Ernst, Thomas. 2014. The syntax of adverbs. In A. Carnie, D. Siddiqi and Y.
Sato (eds.), The Routledge Handbook of Syntax, 108–130. New York: Routledge.
Ernst, Thomas. 2020. The syntax of adverbials. Annual Review of Linguistics 6(1):
89–109.
Hamano, Shoko. 1998. The Sound-Symbolic System of Japanese. Stanford, CA: CSLI
Publications.
Hirose, Masayoshi. 1981. Japanese and English Contrastive Lexicography: the Role
of Japanese ‘Mimetic Adverbs’. Unpublished PhD dissertation, University of
California, Berkeley.
Ibarretxe-Antuñano, Iraide. 2006. Sound Symbolism and Motion in Basque.
Munich: Lincom Europa.
Jackendoff, Ray S. 1972. Semantic Interpretation in Generative Grammar. Cam-

Kulemeka, Andrew Tilimbe Clement. 1993. The Status of the Ideophone in

Chichewa. Unpublished PhD dissertation, Indiana University.
Lakoff, George. 1987. Women, Fire, and Dangerous Things: What Categories Reveal
about the Mind. Chicago: University of Chicago Press.
Martin, Samuel. 1975. A Reference Grammar of Japanese. New Haven, CT: Yale
University Press.
McCawley, James D. 1968. The Phonological Component of a Grammar of Japanese.
The Hague: Mouton.
McConnell-Ginet, Sally. 1982. Adverbs and logical form. Language 58:
144–184.
Moshi, Lioba. 1993. Ideophones in KiVunjo-Chaga. Journal of Linguistic Anthro-
pology 3(2): 185–216.
Newman, Paul. 1968. Ideophones from a syntactic point of view. The Journal of
West African Languages 5: 107–117.
Nuckolls, Janis B. 1996. Sounds Like Life: Sound-Symbolic Grammar, Performance
and Cognition in Pastaza Quechua. New York: Oxford University Press.
Patent, Jason. 1998. A willy-nilly look at Lai ideophones. Linguistics of the
Tibeto-Burman Area 21(1): 155–200.
Reiter, Sabine. 2011. Ideophones in Awetí. Unpublished PhD dissertation, Uni-
versity of Kiel.
Rivero, María-Luisa. 1992. Adverb incorporation and the syntax of adverbs in
modern Greek. Linguistics and Philosophy 15: 289–331.
Sawada, Harumi. 1978. A contrastive study of Japanese and English sentence
adverbials: From the viewpoint of speech act theory [in Japanese]. Gengo
Kenkyu 74: 1–36.
Schachter, Paul. 1985. Parts-of-speech systems. In T. Shopen (ed.), Language
Typology and Syntactic Description, Vol. I: Clause structure, 3–61. Cambridge:
Schaefer, Ronald P. 2001. Ideophonic adverbs and manner gaps in Emai. In E.
F. K. Voeltz and C. Kilian-Hatz (eds.), Ideophones, 339–354. Amsterdam: John
Benjamins.
Smith, Carlota. 1997. The Parameter of Aspect (2nd ed.). Dordrecht: Kluwer
Academic Publishers.
Storch, Anne. 2014. A Grammar of Luwo: An Anthropological Approach. Amster-
Toratani, Kiyoko. 2005. A cognitive approach to mimetic aspect in Japanese.
Proceedings of the Thirty-first Annual Meeting of the Berkeley Linguistics Society,
335–346.
Toratani, Kiyoko. 2007. An RRG analysis of manner adverbial mimetics.
Language and Linguistics 8(1): 311–342.
Toratani, Kiyoko. 2017. The position of to/Ø-marked mimetics in Japanese
sentence structure. In N. Iwasaki, P. Sells and K. Akita (eds.), The Grammar
of Japanese Mimetics: Perspectives from Structure, Acquisition and Translation
(Routledge Studies in East Asian Linguistics), 35–72. London: Routledge.

426 KIYOKO TORATANI
Tsujimura, Natsuko and Masanori Deguchi. 2007. Semantic integration of

mimetics in Japanese. Communication and Linguistics Studies 39(1): The Main
Session: Papers from the Thirty-Ninth Annual Meeting of the Chicago Linguistic
Society, 339–353.
categories in Role and Reference Grammar. In Robert D. Van Valin., Jr.
(ed.), Investigations of the Syntax–Semantics–Pragmatics Interface, 161–178.
Voeltz, Erhard F. K. and Christa Kilian-Hatz (eds.). 2001. Ideophones. Amster-
Watson, Richard L. 2001. A comparison of some Southeast Asian ideophones
with some African ideophones. In E. F. K. Voeltz and C. Kilian-Hatz (eds.),
Ideophones, 385–405. Amsterdam: John Benjamins.
Watters, James K. 2013. Transitivity, constructions, and the projection of
argument structure in RRG. In B. Nolan and E. Diedrichsen (eds.), Linking
Constructions into Functional Linguistics, 23–40. Amsterdam: John Benjamins.
Notes
1 Ideophones refer to a ‘vivid representation of an idea in sound’ (Doke

1935: 118), or ‘[a] member of an open lexical class of marked words that
depict sensory imagery’ (Dingemanse 2019: 16). Although the term ‘ideo-
phone’ is so commonly used for African and Amerindian languages that it
has become a byword, this chapter uses the term ‘mimetics’ to refer to
Japanese, following the tradition of the literature of Japanese linguistics.
2 While some authors contend ideophones are a subset of regular lexical
categories available in the language, such as adverb and verb (e.g. Moshi
1993; Newman 1968), others say they constitute a lexical category of their
own (e.g. Kulemeka 1993; Reiter 2011). I follow McCawley’s (1968) insights
on Japanese mimetics, separating the Japanese lexicon into four strata
(native, Sino-Japanese, loans from other languages, and mimetics). That is,
I consider ideophones to constitute a lexical stratum of the language
whose linguistic materials are recruited to form a unique word group or
category, but they do not constitute a grammatical category, ‘ideophone’,
per se, on a par with the major categories of verb, noun, adjective
and adverb.
3 Status includes clausal negation and epistemic modality. Modality, listed
under core operators in Figure 9.2, includes root modals such as those
expressing ability and obligation.

4 Adverbs are a lexical rather than a grammatical (functional) category, to

which operators refer (cf. Van Valin 2005: 26).
5 The terms ‘subject-oriented’ and ‘speaker-oriented’ are from Jackendoff
(1972).
6 The term ‘hyperonym’ may be spelled ‘hypernym’. The former is used
here following Croft and Cruse (2004).
7 It is worth noting that the rice-beating ideophone in Kisi, pim pim ‘rice
beaten by two or more people’, can co-occur only with cuu ‘to beat or
pound (usually rice) in a mortar with a pestle’ (Childs 1994: 189), with
the former apparently the hyponym of the latter. But since the ideo-
phone affects the interpretation of the number of event participants, it
must be considered a core modifier.
8 The distinction is not always clear-cut. As a basic rule of thumb, it can be
considered a non-hyponym if the mimetic or ideophone can modify a
general activity meaning ‘do something’: yukkuri nanika-o suru ‘do some-
thing slowly’ vs. *musyamusya nanika-o suru ‘do something munch-munch’.
9 This refers only to cases where the ideophone is contained in the periph-
ery, modifying a given layer of the clause, excluding cases where the
ideophone occurs in the pre-detached position, when it functions as a
clausal adverb.
10 When the adverb is accompanied by the focus particle mo as in bukiyooni-
mo ‘clumsily’, this yields only the clausal (evaluative) reading (cf. Sawada
1978).
11 A reviewer of the chapter noted the Tepehua examples in (28) are remin-
iscent of the case of adverb incorporation in Modern Greek (Rivero 1992).

10
Adposition Assignment and
Adpositional Phrase Types
in RRG
Sergio Ibáñez Cerda
10.1 Introduction
This chapter has a twofold aim: to discuss the RRG treatment of adpositional
phrase (AP) types and adpositional assignment (Foley and Van Valin 1984;
Van Valin 1991, 1993, 2005; Van Valin and LaPolla 1997; Jolly 1993) and to
propose a new typology of adpositional phrase types, following Ibáñez Cerda
(2009, 2011). Although we only discuss data from English and Spanish, the
insights and analyses presented here can be broadly applied to other
languages.
The RRG treatment of adpositional systems relies on three principal
assumptions. First, the function of adpositions is comparable to that of
morphological case, in that adpositional assignment may depend on the
semantic role of the argument. In this sense, adpositions can be considered
to be analytic case forms (see Chapter 7). Second, RRG assumes that the
assignment of adpositions is not idiosyncratic and need not be postulated as
part of the lexical entries of verbs. Instead, adpositional assignment follows
systematic rules, which are applied in the linking and, rather than being
based on grammatical relations or phrase structure positions, depend on the
semantic content of the logical structure (LS) of the predicate and on the
semantic/syntactic distinction between direct and oblique core arguments.
The third assumption is the distinction, based on Bresnan (1982), between
predicative and non-predicative adpositions. Non-predicative adpositions
mark verbal arguments: they do not license these arguments and they do
not add any substantive semantic information to the clause. They are a
function of the semantics of the predicate and, thus, they are free-
morphemic case markers assigned by the predicate (e.g. to in Pat gave a book
to Peter). Contrastingly, predicative adpositions are predicates, in that they
contribute substantive semantic information to the clause in which they
occur, both in terms of their own meaning and of the meaning of the
argument that they license. A typical example of a predicative adposition

Adposition Assignment and Adpositional Phrase Types 429
is one that introduces a peripheral locative – or setting – complement (e.g. in

in Julia had dinner in the garden). Importantly, the two types of adposition are
closely related. In fact, both functions, predicative and non-predicative, can
be, and usually are, played by the same forms. Jolly (1993) takes the predica-
tive adposition to be basic and to be stored in the lexicon. Generally,
adpositions will have a case-marking function when their LS is a portion
of the LS of the predicate that licenses the argument they mark. For
example, English predicative in (as in John had dinner in the park) is closely
related to non-predicative in, which marks the locative argument in John
inserted the key in the lock. The LS of predicative in is be-in′ (z, y), while the LS of
insert is [[do′ (x, Ø)] CAUSE [INGR be-in′ (z, y)]]. Thus, a segment of this last LS
is the same as the LS of the preposition, and this is what is behind the
marking of an argument of insert with in. As will be pointed out in Section
10.5, the distinction between predicative and non-predicative adpositions,
along with the semantic distinction between the notions of verbal argument
and free adjunct, and the syntactic difference between those arguments or
adjuncts that can display core features – such as the possibility of function-
ing as controllers or pivots – and those that cannot, will allow us to outline a
system of seven adpositional types.
The chapter is organized as follows: Section 10.2 summarizes the tenets of
RRG which are key for an understanding of adpositions within the frame-
work. Section 10.3 is an introduction to case and adpositions in RRG. The
current RRG account of adpositional assignment is discussed in Section 10.4.
We then propose our own classification of adpositional phrases (Section
10.5). Finally, in Section 10.6, we draw some brief conclusions.
10.2 Relevant Tenets of RRG
As is stated in Van Valin (2005: 4), RRG bases its analysis of clause structure
on two universal semantic distinctions: the one between the predicate and
non-predicating elements and, on the other hand, among the non-
predicating elements, the distinction between arguments and non-argu-
ments (see Figure 10.1).
The distinction between arguments and non-arguments is based on the
idea that the states of affairs represented by the predicates inherently
determine the number and type of participants involved in them. The
nature of the situation conditions the presence of the participants that are
predicate
Non-arguments
arguments
Figure 10.1 Universal oppositions underlying clause structure (Van Valin 2005: 4)

430 SERGIO IBÁÑEZ CERDA
CLAUSE
CORE PERIPHERY
NUCLEUS
Figure 10.2 Components of the layered structure of the clause (Van Valin 2005: 4)
needed in order to make possible the state of affairs. In this view, the
participants inherently required by a predicate are the arguments, while
those that are not are the non-arguments or adjuncts.
Based on these semantic oppositions, RRG proposes the layered structure
of the clause (LSC), which consists of the nucleus, the core and the periphery,
as represented in Figure 10.2.
Similarly to Lexical Functional Grammar (Bresnan 2001) and Functional
Grammar (Dik 1989), RRG assumes that syntax and semantics work in
parallel. There are no derivations, or any deep levels of representation, but
just one single level with two parallel representations.1 There are natural
correspondences between syntax and semantics. To begin with, the semantic
predicate corresponds to the nucleus of the clause in syntax. Following an
iconicity principle (Haiman 1980), arguments, along with the nucleus,
belong to the core, whereas the non-arguments are peripheral elements of
the clause. Depending on their morphosyntactic properties, arguments can
be direct or oblique. The direct core arguments are marked by direct mor-
phological case: nominative and accusative in nominative-accusative
systems, and absolutive and ergative in absolutive-ergative systems. Periph-
eral non-arguments are adjuncts. Against this backdrop, a clause like John
gave a book to Mary in the library can be represented as in Figure 10.3.
CLAUSE
CORE PERIPHERY
John gave a book to Mary in the library
NU
NUCLEUS
Figure 10.3 Layered structure of John gave a book to Mary in the library
Using the notions just outlined, we can say that John and a book are direct
core arguments, while to Mary is an oblique core argument. The three of
them are participants semantically required by the verb to give. As for the
library, it is an adjunct. Table 10.1 summarizes the relation between the
units of the syntactic and semantic representations.
A third type of clausal component, originally proposed by Jolly (1993), is
that of argument-adjuncts in the core. This complement type corresponds to
APs that code a verbal argument but, at the same time, are introduced by a
predicative preposition.

Table 10.1 Semantic units underlying the syntactic units of the layered structure of
the clause (Van Valin 2005: 5)
S EMANTIC UNIT S YNTACTIC UNIT
Predicate Nucleus
Argument in semantic
representation of predicate Core argument
Non-argument Periphery
Predicate þ Arguments Core
Predicate þ Arguments þ
Non-arguments Clause (¼ Core þ Periphery)
The semantic representation of the clause is built upon the semantic

representation of the verb or the predicating element; it starts from six
basic Aktionsart types (see (1)) and their causative counterparts (2).
(1) a. State: The boy is afraid.

b. Achievement: The balloon popped.
c. Semelfactive: The pencil tapped on the table.
d. Accomplishment: The ice melted.
e. Activity: The soldiers marched in the park.
f. Active accomplishment: The soldiers marched to the park.
(2) a. Causative state: The dog frightens/scares the boy.

b. Causative achievement: The cat popped the balloon.
c. Causative semelfactive: The teacher tapped the pencil on the table.
d. Causative accomplishment: The hot water melted the ice.
e. Causative activity: The sergeant marched the soldiers in
the park.
f. Causative active The sergeant marched the soldiers to
accomplishment: the park.
RRG uses a modified version of the representational scheme proposed in

Dowty (1979) to capture these classes (see Table 10.2).
Table 10.2 Lexical representations for Aktionsart classes
Aktionsart class Logical structure
STATE predicate′ (x) or (x, y)

ACTIVITY do′ (x, [predicate′ (x) or (x, y)])
ACHIEVEMENT INGR predicate′ (x) or (x, y), or
INGR do′ (x, [predicate′ (x) or (x, y)])
SEMELFACTIVE SEML predicate′ (x) or (x, y)
SEML do′ (x, [predicate′ (x) or (x, y)])
ACCOMPLISHMENT BECOME predicate′ (x) or (x, y), or
BECOME do′ (x, [predicate′ (x) or (x, y)])
ACTIVE ACCOMPLISHMENT do′ (x, [predicate1′ (x, (y))]) & INGR predicate2′ (z, x) or (y)
CAUSATIVE α CAUSE β, where α, β are logical structures of any type

An assumption which distinguishes RRG from all other syntactic frame-

works is that it posits two types of semantic role. The traditional thematic
relations of agent, patient, theme, instrument, etc. are only mnemonics for five
structural positions in the LS of predicates. In addition to these five positions,
RRG postulates the two semantic macroroles actor and undergoer, which are
generalizations across thematic relations. Actor is a generalization across
agent, experiencer, instrument, and other roles, while undergoer is a general-
ization subsuming patient, theme, recipient, and other roles. Agent is the
prototype for actor, and patient is the prototype for undergoer. The two
macroroles are the two primary arguments of a transitive predication, and
either one of them can be the single argument of an intransitive verb.
The logical structure of the predicate determines which macroroles it
takes. If it takes two macroroles, then they must be actor and undergoer.
For verbs which have a single macrorole, the default choice follows directly
from the logical structure of the verb: if the verb has an activity predicate in
its logical structure, the macrorole will be actor; otherwise, it will be under-
goer. The default macrorole assignment principles are summarized in (3)
(Van Valin 2005: 63).

number of arguments in its logical structure.
1. If a verb has two or more arguments in its logical structure, it will take
two macroroles;
2. If a verb has one argument in its logical structure, it will take one
macrorole.
b. Nature: for verbs which take one macrorole,
1. If the verb has an activity predicate in its logical structure, the
macrorole is actor.
2. If the verb has no activity predicate in its logical structure, the
macrorole is undergoer.
Although most verbs follow these defaults assignments, there are excep-
tions (e.g. intransitive two-place predicates such as locative or psych verbs).
These must be specified in the lexical entries of the relevant verbs in terms
of a simple feature [MR α], with values [MR 0], [MR 1] and [MR 2].
Finally, the relation between macroroles and logical structure argument
positions is captured in the Actor-Undergoer Hierarchy (AUH) in Figure 10.4
(Van Valin 2005: 61).
ACTOR UNDERGOER
Arg. of 1st arg. of 1st arg. of 2nd arg. of Arg. of

DO do′ (x,... pred′ (x, y) pred′ (x, y) pred′ (x)
[ = increasing markedness of realization of argument as macrorole]
Figure 10.4 Actor-Undergoer Hierarchy

Given the logical structure of a transitive verb, the leftmost argument will
by default be the actor and the rightmost argument will be the undergoer.
Marked assignments to undergoer are possible, typically with three-place
predicates, where there can be two arguments competing for a macrorole
function, as in (4a–c).
(4) a. [do′ (Pat, Ø)] CAUSE [BECOME have′ (Chris, book)]

b. Pat [actor] gave the book [undergoer] to Chris.
c. Pat [actor] gave Chris [undergoer] the book.
Example (4c) illustrates the dative shift alternation, whereby the first argu-
ment of the two-place state predicate, not the second, is assigned the
macrorole undergoer.
10.3 Adpositions in RRG
10.3.1 Case and Adpositions

‘Case is a system of marking dependent nouns for the type of relationship
they bear to their heads’ (Blake 2001: 1). Its primary function is to establish
and to differentiate, in semantic and grammatical terms, the role or func-
tion of verbal arguments with respect to the event denoted by the verb. RRG
differs from other syntactic theories, where case marking and agreement
are invariably tied to grammatical relations. Since notions like subject and
direct object have no place in this framework (see Chapter 5), neither case
marking nor agreement can be based on them. Rather, case-marking rules
make crucial reference to macroroles and direct core argument status, in
addition to the content of the LS in the lexical entry of the verb, as we will
show in Section 10.4.2 for oblique case adpositions. Grammatical aspect,
modality and negation can also play a role in case marking. For example, in
Slavic languages, the direct object of negated clauses is marked by genitive
instead of accusative case. In addition, in Spanish and other languages, the
animacy or inherent lexical content of a reference phrase (RP) (Silverstein
1976, 1981, 1993) also plays a role in case marking. For example, dative
appears instead of (al)lative or genitive case in many constructions to mark
nominals whose referent is animate.
RRG posits a set of rules for direct or grammatical case assignment (see
Chapter 7). Grammatical or direct cases are not normally marked by adposi-
tions and can have an agreement cross-reference on the verbal head: nom-
inative, accusative and dative in accusative languages, and absolutive,
ergative and dative in ergative languages. However, in some languages,
direct cases can be instantiated by adpositions, like the a preposition for
dative case in Spanish, as in Román dio una flor a Tere ‘Román gave a flower to
Tere’. Besides the morphological case forms, many languages have a comple-
mentary system of adpositions. In addition, in some languages, such as Latin
and other Indo-European languages, adpositions can govern specific case

forms, although this type of case marking goes beyond the scope of
this chapter.
In RRG, the notion of case marking is extended to cover the function of the
cross-referencing pronominal affixes which appear on the verbal base in
head-marking languages (Nichols 1986, Van Valin 2005: 16–17). These affixes
perform a twofold function: on the one hand, they indicate the relation
between the arguments and the verb, and on the other hand, they are
arguments in syntax. Only the grammatical or direct cases are marked on
the head verb; other semantic relations, if marked, will be marked through
adpositions or morphological case on the dependents.
10.3.2 Difference between Predicative and Non-Predicative

Adpositions
Building upon Bresnan (1982), RRG posits two types of adposition: predica-
tive and non-predicative. Predicative adpositions function as predicates in
that they introduce a participant which is not licensed by the verb. This
participant is their object, and they are the head of the phrase, which is an
adjunct. Non-predicative prepositions, on the other hand, mark an argu-
ment licensed by the verb. Thus, they are a function of the semantics of the
verb; they themselves do not add any semantic information to the clause.
They are like free-morphemic case markers assigned by the predicate. In John
gave a book to Mary in the library (see Figure 10.3), to Mary is a non-predicative
prepositional phrase (PP) functioning as a core argument, while in the library
is a predicative PP functioning as an adjunct.
The two types of adposition are represented differently in the layered
structure of the clause and hence in the constituent projection. Since non-
predicative adpositions mark arguments of the verb, the phrases in which
they occur are APs in the core; the adposition is treated as a case marker, and
it is not the head of the phrase. Predicative adpositions, on the other hand,
have their own LS and are the head of the phrase in which they occur. The
nominal they introduce is their object or argument. The structure of a non-
predicative or argument AP and a predicative AP is given in Figure 10.5 and
Figure 10.6, respectively.
As mentioned before, there is a third function that an adposition can
have: it can mark an argument of the verb, while at the same time contrib-
uting its semantics to the clause. This type of PP typically, but not exclu-
sively, appears with locative verbs like those that denote a change of place
(e.g. put). Such verbs require a goal argument but the choice of the locative
adposition is not completely determined by the verb, as the speaker can
choose among some very closely related adpositions, which semantically
elaborate on the type of locative relation between the theme and the goal, as
shown in (5).
(5) Kim put the book {in / on / next to / behind / on top of / under} the box.

PP
COREP
PP
NUCP RP
PRED
P RP
P
to Pat in the library
Figure 10.5 English non-predicative PP Figure 10.6 English predicative PP
In this example, the different prepositions contribute an important com-

ponent of meaning, unlike the preposition to with a verb like give or show,
and therefore they must be considered to be predicative prepositions. As
such, they have the structure of a predicative AP (see Figure 10.5), although
they occur in the core, as they mark a verbal argument. In RRG terms, they
are ‘argument-adjuncts’ in the core. In Section 10.5 we propose a typology
of adpositional phrases in terms of their different semantic and
syntactic status.
10.4 Adpositional Phrase Types and Adposition

Assignment in RRG
In this section, we discuss the RRG treatment of the three types of adposi-
tional phrase introduced above: predicative (10.4.1), non-predicative (10.4.2)
and argument-adjunct (10.4.3), drawing on Jolly (1993), Van Valin and
LaPolla (1997) and Van Valin (2005).
10.4.1 Predicative or Adjunct Adpositional Phrases

As was mentioned, predicative prepositions introduce a participant which is
not an argument in the LS of the main verb, that is, an adjunct. They are
treated as primitives, in that they have their own lexical entry in the
lexicon, and the syntactic phrases of which they are heads have the status
of modifiers in the periphery. There are three main types of predicative
adposition: (a) locative and temporal setting adpositions (e.g. in the park in
(6a) below); (b) sequential adpositions (before, after and during/while); and
(c) causative adpositions (e.g. English for). In the standard RRG treatment,
predicative adpositions are taken to be two-place predicates which take the

LS of the nuclear verb of the clause as one of their arguments. This is

illustrated in (6a–b).
(6) a. Chris ran in the park.

b. be-in′ (park, [do′ (Chris, [run′ (Chris)])])
c. Chris ran to the park.
d. (do′ (Chris, [run′ (Chris)]) & INGR be-at′ (park, Chris))
In (6a), Chris’s running takes place in the park, and therefore the logical
structure of the predicative preposition in is the highest predicate in the
logical structure; it takes the park and the logical structure for run as its two
arguments. This contrasts with the logical structure of a clause like (6c) Chris
ran to the park, where the PP expresses the location of the referent of Chris,
not the location of the event of running, and in this example to the park is an
argument-adjunct PP (see Section 10.4.2).
As pointed out in Ibáñez Cerda (2009), predicative or setting PPs are like
adverbs, in that they can modify different parts of a verbal LS, and not
necessarily all of it, as shown in (7) for Spanish.
(7) Juan cortó el pastel sobre la tabla de madera.

John cut the cake on the board of wood
‘John cut the cake on the wooden board.’
Here the PP sobre la tabla de madera ‘on the wooden board’ only refers to the
place where the change of state takes place; more precisely, the wooden
board is the place where the event of ‘dividing the cake into pieces’ happens.
Certainly, Juan is doing something that involves the wooden board, but,
clearly, he is not located on the board. Consequently, the semantic scope of
the PP is not over the core but only part of it. In this way, we can consider
PPs like the one in (7) as partial modifiers of the core. A possible representa-
tion for (7) is thus shown in (8).
(8) [do′ (Juan, Ø)] CAUSE [be-on′ (tabla, [BECOME cut′ (pastel)])]2
In contrast to the LS in (6b), here the adverbial predicate be on′ only has
scope over the subevent of change of state, leaving the activity subevent out.
Apart from the case of (7), the fact that adjunct PPs can modify structural
elements other than the whole core can be seen in a sentence with an added
instrumental complement (9).
(9) Juan cortó el pastel con un cuchillo sobre la tabla de madera.

John cut the cake with a knife on the board of wood
‘John cut the cake with a knife on the wooden board.’
Not only is the change of state under the scope of the locative PP, but
part of the action performed by the effector falls inside its scope: John’s
knife manipulation and, more clearly, the contact of this instrument
with the cake, takes place on the wooden board. This can be represented
as follows.

(10) [do′ (Juan, use′ [Juan, cuchillo])] CAUSE [[be-on′ (tabla, [do′ (cuchillo, [cut′
(cuchillo, pastel)])])] CAUSE [BECOME cut′ (pastel)]]
Here, only one of the activity subevents, the one that has the instrument as
an effector, is under the scope of be-on′, while the more external one, the
one where Juan is the effector, falls out of the scope of the preposition.
In sum, following Ibáñez Cerda (2009), we have argued that the PPs in (7)
and (9) behave differently from the one in (6a): the whole core of the clause is
one of the arguments of the PP in (6a), but this is not the case with (7) and (9).
Nevertheless, none of these PPs codify verbal arguments, they are all headed
by a predicative preposition, and they are all modifiers in the periphery. In
this sense, they represent different cases of adjunct complements intro-
duced by predicative PPs.
10.4.2 Non-Predicative Adposition Assignment and Oblique

Core Arguments
Non-predicative adpositions mark a verbal argument: specifically, in RRG
terms, an oblique core argument. As was mentioned, these adpositions are
not idiosyncratically listed in the lexical entries of verbs, but rather assigned
in terms of systematic rules (Foley and Van Valin 1984; Jolly 1993; Van Valin
and LaPolla 1997). The basic rules governing the assignment of to, from and
with in English are given here.
(11) a. Assign to to non-MR y-argument in logical structure segment:

BECOME/INGR pred′ (y, z)
b. Assign from to non-MR y-argument in logical structure segment:
BECOME/INGR NOT pred′ (y, z)
c. Assign with to non-MR b argument if, given two arguments, a and b, in a
logical structure with (i) both as possible candidates for a particular
macrorole and (ii) a being equal or higher (to the left of b) on the AUH, b is
not selected as that macrorole.
We provide some examples of the assignment of the English preposition to

in (12a–c). Then, in (13) we illustrate the relevant LS segments.
(12) a. Sally gave/sent/handed the box to Pat.

b. Sally showed the box to Pat.
c. Sally taught basketweaving to Pat.
(13) a. . . .BECOME have′ (Pat, box) ¼ give, hand, (send)

b. . . .BECOME see′ (Pat, box) ¼ show
c. . . .BECOME know′ (Pat, basketweaving) ¼ teach
In all these examples, the RP marked by to is the first argument of a two-

place predicate embedded under a BECOME operator, and it is also a non-
macrorole core argument. The state predicate can be a possession (12a, 13a),
perception (12b, 13b) or cognition verb (12c, 13c), as well as a locative
predicate (as with send in 12a). Therefore, the argument marked by to can

be a possessor, a perceiver, a cognizer or a location. Although to marks

different types of semantic argument, it always appears in the same kind
of LS environment, precisely the one foreseen in the rule in (11a): BECOME/
INGR pred′ (y, z)
A similar analysis can be posited for from, which appears in the examples
in (14a–c), the relevant LS segments being given in (15a–c).
(14) a. Sandy took/stole/bought the keys from Kim.

b. Pat drained the water from the pool.
c. Kim escaped from the burning house.
(15) a. . . . BECOME NOT have′ (Kim, keys) ¼ take, steal, buy

b. . . . BECOME NOT be-in′ (pool, water) ¼ drain
c. . . . BECOME NOT be-in′ (burning house, Kim) ¼ escape
In each of the LS segments in (15a–c), from marks the first argument of the
two-place state predicate, which is a non-macrorole core argument. The
difference between these segments and those in (13a–c) is the presence of
NOT. This difference in content between to and from was first proposed by
Gruber (1965). Again, like to, from does not mark a single thematic relation,
but rather it is assigned in a particular LS context – BECOME/INGR NOT pred′
(y, z), and the rule (11b) rightly predicts its appearance.
The rule (11c) for preposition with applies in the single-prime examples in
(16a′, b′), where the theme argument z is not selected as undergoer and is
instead marked by with. In contrast, the non-prime examples (16a, b) obey
rule (11a), which requires the marking with to.
(16) a. Sally presented the flowers [z] to Kim [y].

a′. Sally presented Kim [y] with the flowers [z].
a′′. [do′ (Sally, Ø)] CAUSE [BECOME have′ (Kim, flowers)]
b. Max loaded the olives [z] into his minivan [y].
b′. Max loaded his minivan [y] with the olives [z].
b′′. [do′ (Max, Ø)] CAUSE [BECOME be-in′ (minivan, olives)]
Thus, with is assigned in contexts where two arguments can be selected for
the undergoer status, but the marked option – in terms of (11a) – is the one
that prevails.
There are two further contexts where with is assigned: in the marking of
an instrument (17a) and in a comitative PP (17b).
(17) a. Tom cut the bread with the knife

a′. The knife cut the bread
a′′. [do′ (Tom, [use′ (Tom, knife)])] CAUSE [[do′ (knife, Ø)] CAUSE [BECOME cut′
(bread)]]
b. Sandy went to the store with Kim
b′. Sandy and Kim went to the store
b′′. [do′ (Sandy/Kim, [move′ (Sandy/Kim)]) ^ PROC cover.path.distance′ (Sandy/
Kim)] & INGR be-LOC′ (store, Sandy/Kim)

In (17a), following the AUH, it is the leftmost argument, Tom, which is

assigned the macrorole actor, and as such it is projected as subject. The
other argument, the instrument, the knife, is marked by the preposition with.
However, when the effector is not projected, the instrument can be selected
as actor and then projected in the subject function, as can be seen in (17a′).
A similar alternation is found with the comitative. The LS in (17b′′) has two
co-agent arguments, Sandy and Kim, and each one can be assigned the
macrorole actor and be projected as subject. In (17b) Sandy is projected as
subject and, applying the rule (11c), Kim is marked by the preposition with;
but in (17b′) both referents are arguments as subject, through a single
complex RP.
Summing up, English with appears in contexts where two arguments
compete for a macrorole status, actor or undergoer, and one is not selected
as such, the context foreseen by the rule in (11c).
In languages with morphological case (see Chapter 7), dative is usually
assigned to mark the recipient of transfer verbs, like those in (12a–c). This
marking obeys what is considered in RRG (Van Valin and LaPolla 1997) to be
the default rule for the assignment of oblique cases.
(18) Assign dative case to non-macrorole arguments (default).
Thus, in a Croatian example like (19), unuc- ‘grandsons’, the highest-

ranking argument, is selected as actor; cvijet- ‘flowers’, the lowest-ranking
argument, is selected as undergoer, and bak- ‘grandmother’, the non-
macrorole core argument, receives dative case.
(19) Unuc-i su bak-i darova-l-i cvije′c-e.

grandson-m.pl.nom be.3pl grandmother-f.sg.dat give-pst-pl flower-m.pl.acc
‘The grandsons gave flowers to [their] grandmother.’
As seen above, languages which do not have morphological case

marking on nouns use adpositions to mark oblique case. While English
uses the preposition to, Spanish uses the preposition a, which is the same
form as appears with the goal argument of motion verbs. The difference
between the goal preposition and the dative one is that the complement
marked by dative a is replaceable or commutable for the dative clitic
pronoun le/les; indeed, it usually appears reduplicated or cross-referenced
by this clitic.
This same preposition a, cross-referenced by the pronoun le/les, also
appears with M-intransitive verbs, like the psychological ones gustar ‘to like’
and encantar ‘to love’, as in A María le gusta el chocolate ‘Mary likes chocolate’.
These are two-place predicates with only one macrorole argument, which is
what the M-intransitive specification means. Since they are also state predi-
cates – like′ (María, chocolate) – their only macrorole argument must be an
undergoer and, in terms of the AUH, the second argument outranks the first
for the undergoer status. The undergoer receives nominative case, while the

non-macrorole argument is assigned dative case, which is the default

oblique case.
Besides psych verbs, there is, however, another large class of M-intransitive
predicates in Spanish, whose non-macrorole argument is marked by the
preposition de ‘of’, which is the form which also marks the genitive and
the ablative. These are verbs like carecer ‘to lack’, constar ‘to consist’, desconfiar
‘to distrust’, abusar ‘to abuse’ and prescindir ‘to dispense’, which, according to
the literature (Alarcos Llorach 1972; Martínez García 1986; Hernández
Alonso 1990; Cano Aguilar 1999; among others), govern their preposition,
which basically means that they have their specific prepositional mark
idiosyncratically specified in their lexical entry. Given that the presence of
the preposition de is systematic with the majority of these verbs, one can
consider them as M-intransitive and, then, postulate the application of a
general rule for the assignment of the genitive preposition. This would need,
however, an extra specification U¼x in the verbal entries, to signal that a
marked choice of undergoer must take place, given that in these cases the
first argument is selected as undergoer, contrary to the specifications of
the AUH. In this way, in an example like (20a), it is María which gets the
macrorole status and the nominative case, and oportunidades is marked by
the genitive case.
(20) a. María carece de oportunidades.

Mary lacks gen opportunities
‘Mary lacks opportunities.’
b. lack′ (María, oportunidades)
c. lack′ (x, y) U¼x
In Spanish, then, there are two clearly identifiable classes of M-

intransitive verbs: one marked by dative case and another one marked by
genitive case. Thus, the Spanish non-macrorole assignment rule can be
reformulated as (21).
(21) a. Assign dative case to non-macrorole arguments (default).

b. Assign genitive case to non-macrorole arguments of verbs with the
specification U¼x
The importance of the genitive adposition as a non-semantic or gram-

maticalized mark for non-macrorole arguments in Spanish is additionally
attested when considering its ‘competing’ role for marking arguments
that are not selected for macrorole status, in contexts where those argu-
ments could or should have been picked up as such. These are contexts
where, in English, the rule for with would apply. This is the case with the
oblique argument of the transfer (16a′) and locative (16b′) alternation
construction. Although some constructions of this type, where a non-
default assignment has taken place, can be marked by the preposition
con, the equivalent of with (see, e.g. 22e′, f′), the genitive preposition de
appears in all of them.

(22) a. La oficina de prensa informó la noticiaal público.

The office of press informed the notice dat.def.art public
‘The press office gave the news to the public.’
a′. La oficina informó al público de /*con la noticia.
The office informed acc.def.art public gen/*ins the notice
‘The press office informed the public the notice.’
b. Elia le perdonó su mala educación a Marcelo.
Elia dat forgave his bad education dat Marcelo
‘Elia forgave his bad manners to Marcelo.’
b′. Elia perdonó a Marcelo de /*con su mala educación.
Elia forgave acc Marcelo gen/*ins his bad education
‘Elia forgave Marcelo his bad manners.’
c. Ramón denunció el robo de Martha.
Ramón reported the robbery of Martha
‘Ramón reported Martha’s robbery’.
c′. Ramón denunció a Martha de /*con robo.
Ramón reported acc Martha gen/*ins robbery
‘Ramón reported Martha for the robbery’.
d. Ludmila vació todo el líquido de la botella.
Ludmila emptied all the liquid from the bottle
‘Ludmila emptied all the liquid from the bottle.’
d′. Ludmila vació la botella de /*con todo su líquido.
Ludmila emptied the bottle gen/*ins all its liquid
‘Ludmila emptied the bottle of all its liquid.’
e. Leonardo cargó las manzanas en el camión.
Leonardo loaded the apples in the truck
‘Leonardo loaded the apples into the truck.’
e′. Leonardo cargó el camión de/ con manzanas.
Leonardo loaded the truck gen/ins apples
‘Leonardo loaded the truck with the apples.’
f. Fernando roció cerveza en todo el cuarto.
Fernando sprayed beer in all the room
‘Fernando sprayed beer all over the room.’
f ′. Fernando roció el cuarto de /con cerveza.
Fernando sprayed the room gen/ins beer
‘Fernando sprayed the room with beer.’
As can be seen from the prime examples in (22), de ‘of’, is the preposition
that, in effect, can appear in all contexts of non-default projection of a
theme, when the competition for the undergoer function is between asym-
metrical arguments, that is, two arguments which do not have the same
status in LS. The domain of con ‘with’ is instead much narrower.
One way of analysing these facts is to consider de to be the basic prepos-
ition for marking non-prototypical argument projections in Spanish. In this
analysis, Spanish de – including the one with genitive function – is an empty
preposition which marks asymmetric relations in non-default cases: (i) lexic-
ally asymmetrical relations between the two arguments of M-intransitive
verbs with the U¼x specification; (ii) non-default coding of asymmetrical

arguments; and (iii) the asymmetrical relations between a noun and its
modifiers. We can thus posit the rule in (23) for the assignment of the de
preposition to verbal arguments in Spanish.
(23) Assign de to non-macrorole arguments in non-default projections.
The dative preposition a can still be the default preposition for non-
macrorole arguments, whereas con ‘with’ is the preposition for: (i) the non-
actor co-agent in comitative constructions (e.g. Lola fue al cine con Domingo
‘Lola went to the movies with Domingo’); (ii) the non-actor instrument of cut
and break verbs; and (c) the non-default coding of arguments with locative
and spray verbs.3
Note that the instrumental case rule in (11c) can still partially apply for
the assignment of the preposition con in Spanish.
(24) Assign con to non-MR b argument if, given two arguments, a and b, in a
logical structure with (i) both as possible candidates for a particular
macrorole and (ii) a being equal or higher (to the left of b) on the AUH, b is
not selected as that macrorole.
This rule effectively covers the cases where con competes with de for the
marking of theme arguments not selected as undergoer (locative and spray
verbs) and cases where de does not appear: for example, those where there
are two arguments competing for the actor macrorole (comitative and
instrument cases). De, on the other hand, appears in marked cases where a
theme argument is not selected as undergoer. Thus, the rule in (23) must be
revised as in (25).
(25) Assign de to non-macrorole arguments in non-default undergoer selections.
Although both (24) and (25) apply with locative and spray verbs, elsewhere
they have their own niche of functionality.
10.4.3 Argument-Adjuncts
Currently, two kinds of argument-adjunct adpositional phrases are recog-
nized in RRG: (a) those that are headed by an adposition which has semantic
content but nonetheless marks a verbal argument, and (b) those phrases
headed by a predicative adposition which introduces a participant not
licensed by the verb (i.e. an adjunct) but does not take the whole LS of the
verb as an argument, sharing its argument with the verbal LS. Both types are
core phrases in syntactic terms.
A typical example of the first type of argument-adjunct is the variable
preposition which codes the goal argument of motion and change-of-place
verbs. Spanish examples are provided in (26) (see also the English example
in (5)).

(26) a. Juan fue a / hacia / hasta / adentro de la tienda.

John went to / towards / as far as / inside of the house.
‘John went to / towards /as far as / inside the house.’
b. Juan puso el libro en /sobre / dentro de / debajo de la caja.
John put the book in-on /over / inside of / under of the box
‘John put the book in-on /over / inside / under the box.’
Ibáñez Cerda (2009) points out that these PPs have different status
depending on the preposition that introduces them. In Spanish, if a goal
argument is coded by a PP introduced by the preposition a ‘to’, with intransi-
tive motion verbs, and by en ‘in/on’, with change-of-place verbs, it should be
analysed as an oblique core argument. Indeed, these prepositions are canon-
ical with the said verb classes. A corpus-based study (Ibáñez Cerda 2005)
shows that the goal PP of intransitive motion verbs was coded with a in over
90 per cent of cases. Similarly, the PP of the change-of-place verbs strongly
tended to be introduced by en. These prepositions are thus assigned system-
atically as the unmarked prepositions in the following contexts: a appears
when the LS of a predicate has a BECOME pred′ (z, y) segment, which is the
case of motion verbs, and en shows up in the structural environment
characterized by the presence of INGR pred′ (z, y), which characterizes
‘putting’ verbs.
By contrast, when the goal arguments are introduced by other prepos-
itions that have more semantic content, such as dentro ‘inside’, hacia
‘towards’, etc., they should be considered as argument-adjuncts in the core,
because although these kinds of prepositions are clearly predicative, the
possibility of using them for introducing the goal argument of change-of-
place verbs is not entirely free, as is the case of the locative PPs that do have
the status of adjuncts. Poner ‘to put’, the predicate that serves as nucleus in
the sentences exemplified in (26b), is the hypernym in the domain of
change-of-place verbs. As such, it has a very general and abstract locative
meaning; it does not inherently specify much about its goal, and that is
why it can be used with goals introduced by almost any locative prepos-
ition. This is not the case, however, with other verbs whose inherent
semantics blocks the use of certain prepositions for coding their goals, as
can be seen in (27).
(27) a. ??Juan metió el libro fuera de la caja.

John put.inside the book out of the box
‘John put inside the book out of the box.’
b. ??Juan sacó el libro dentro de la caja.
John took.out the book inside of the box
‘John took the book out inside the box.’
c. ?? Juan encerró su perro hacia su casa.
John locked his dog towards his house
‘John locked his dog towards his house.’

d. ?? Juan hospedó a Pedro fuera de su casa.

John lodged acc Peter out of his house
‘John lodged Peter out of his house.’
e. ??Juan clavó el clavo detrás de la pared.
John nailed the nail behind of the wall
‘John nailed the nail behind the wall.’
f. ??Juan sumergió la cabeza debajo de la fuente.
John submerged the head under of the fountain
‘John submerged his head under the fountain.’
The examples in (27) indicate that the preposition is determined by the

predicate it appears with, and hence, it is a function of that predicate,
although it can in turn contribute a portion of meaning to the
argument.
Importantly, the PPs with non-canonical prepositions behave as core argu-
ments. They control pivots in coordinated clauses, which in RRG is a stand-
ard diagnostic for core participants. Thus, the most straightforward reading
of (28a) is one where it is the cabin that looks fine. Similarly, (28b) would
normally be taken to mean that the fishbowl looks good.
(28) a. Paseando por el bosque, Juan i llegó hasta la vieja cabañaj

Walking through the forest John arrived to.limit the old cabin
y aún i/j se veía bien.
and still refl looked fine
‘Walking in the forest, John came across the old cabin and (it) looked fine.’
b. Juan i puso un pezh dentro de la pecera nuevaj y *i/ h/j se ve bien.
John put a fish inside of the fishbowl new and refl looks good
‘John put a fish inside the new fishbowl and (it) looks good.’
The LS representation of clauses with these types of argument-adjunct

phrase is not different from those with a goal oblique core argument (cf.
(13)). The argumental value of the goal is guaranteed by the BECOME pred′
(z, y) segment of the LS of the verb. What changes is the LS of the preposition
which fills the pred′ function and instantiates the semantic value which
that form has in the lexicon.
The second type of argument-adjunct is best exemplified by benefactive
complements, like for Sandy in (29).
(29) Robyn baked a cake for Sandy.
Here, for Sandy is an argument-adjunct because the LS of the predicative

preposition that heads it, for, has an argument that is also an argument of
the main predicate, as can be seen in the representation in (30), proposed by
Van Valin and LaPolla (1997: 383).
(30) [[do′ (Robin, Ø)] CAUSE [BECOME baked′ (cake)]] PURP [BECOME have′
(Sandy, cake)]

In (30), the PP for Sandy is represented by the segment introduced by the

PURP operator. This introduces a substructure with two arguments: Sandy,
the beneficiary, is the argument of the preposition; the other argument,
cake, is shared with the LS of the verb. The sharing of an argument with the
verb differentiates this kind of PP from the adjunct ones, and it is because of
this sharing that they are argument-adjuncts in the core.
We propose here, however, that there are differences between the two
types of argument-adjunct introduced above. The goal of motion and
change-of-place verbs, which can be introduced by variable prepositions,
are arguments of the verb and, as shown above, can function as controllers
of pivots. Thus, they are proper arguments-adjunct in the core. The bene-
factive, instead, can hardly be said to be a notion that is inherently linked,
in semantic terms, to a particular state of affairs. Although it usually
appears in clauses where the main verb is an accomplishment predicate
(Jolly 1993), beyond this, there is no other clue for predicting its appearance
with particular semantic classes of predicates. Moreover, benefactives
cannot control pivots in coordinated clauses, as is suggested by the Spanish
example in (31). As noted above, this behaviour is a standard test for
identifying arguments that belong to the core.
(31) Juani hizo un traje para Pedro j y lo i /??j usó todo un día.
John made a suit for Peter and acc.3sg wore all one day
‘John made a suit for Peter and (he) wore it one complete day.’
Even though the suit was made for Pedro, the most natural reading of this
example is that John was the one who wore it for a day (perhaps to check
how it looked on him before giving it to Pedro).
These facts suggest that, in marked contrast with the goal arguments of
motion and change-of-place verbs, benefactives introduced by para are not
core arguments. This suggestion is further supported by the fact that Span-
ish has an alternative mechanism for making core arguments out of bene-
factives; this is the dative or indirect object construction, where the
beneficiary is introduced by the preposition a, and in which it can be
doubled by the clitic pronoun le. We refer to Ibáñez Cerda (2009) for more
in-depth discussion on the different behaviour of benefactives, on the one
hand, and datives and recipients, on the other.
Another independent criterion to determine that the PPs introduced by
para in Spanish do not belong to the core is that there are no non-predicative
uses of this preposition. The Spanish counterparts of the English verbs to long
and to hope, which appear with non-predicative for, are mostly transitive (e.g.
espero las buenas nuevas ‘I’m waiting/hoping for good news’). When they are
used intransitively, they are coded with the preposition por, and not with
para (e.g. espero por la buena nueva).
In sum, PPs introduced by para in Spanish are always predicative and do
not behave as core arguments: they code a participant that is not part of the
verb semantics. Nevertheless, they are not like the adjunct PPs analysed in

Section 10.4.1, because they share one argument with the LS of the verb and,
although they are introduced by a predicative preposition, they do not take
the whole clause as one of their arguments. Our proposal is that they are
argument-adjuncts in the periphery.
As pointed out by Jolly (1993) and Van Valin and LaPolla (1997), the
preposition for is a good example of an adposition which can have the three
main functions discussed above: it can be a non-predicative case marker in
an oblique argument PP (32a), it can function as an argument-adjunct, as in
(29), which is repeated in (32b) for convenience, and it can appear as a
predicative preposition heading an adjunct phrase (32c).
(32) a. Lucy longs for a diamond ring.

b. Robyn baked a cake for Sandy.
c. Rita sang for the students.
As a way of unifying these three uses, Jolly (1993) proposes this extended
semantic representation, which is abbreviated as PURP [. . .] in the logical
structure in (30).
(33) want′ (x, LS) ^ DO (x, [LS1. . .CAUSE. . .LS2])
This can be paraphrased as ‘x wants some state of affairs (described in LS2) to

obtain, and intentionally does LS1, in order to make LS2 happen’. The DO
implies that the action in LS1 cannot be non-volitional. In the case-marking
function of the preposition, only the first part of (33) gets projected. Lucy
longs for a diamond ring can be also paraphrased as ‘Lucy longs to have a
diamond ring’, so for really stands for a reduced proposition. Thus, the LS of
(32a) is want′ (Lucy, [have′ (Lucy, diamond ring)]), a representation that
exactly matches the first part of (33).
As seen above, the argument-adjunct use implies the projection of the
whole LS in (33). The complete representation of Robyn baked a cake for Sandy
is thus as follows.
(34) [want′ (Robyn, [BECOME have′ (Sandy, cake)])] ^ [[DO (Robyn, [do′ (Robyn, Ø)]
CAUSE [BECOME baked′ (cake)]] CAUSE [BECOME have′ (Sandy, cake)])]
The representation of the predicative use of for in (32c) is very similar to

that in (34), except that there is no sharing of arguments between LS1 and
LS2. The benefactive the students, the sole argument of LS2, is introduced as a
new argument in LS1.
(35) [want′ (Rita, [BECOME entertained′ (the students)])] ^ [[DO (Rita, [do′ (Rita,
[sing′ (Rita)])])] CAUSE [BECOME entertained′ (the students)]]
10.5 A New Typology of Adpositional Phrases in RRG
The current RRG three-way distinction between oblique core arguments,

peripheral adjuncts and argument-adjuncts in the core has been very

helpful in overcoming the limitations of the traditional binary distinction

between arguments and adjuncts. However, on closer examination, it
appears that adpositional phrases can play several different functions, in
terms of their relation to the clause they appear in. In fact, in Section 10.4.3,
we argued that the argument-adjunct status can be ascribed to prepositional
phrases with both core and peripheral behaviour, giving rise to a new
distinction between argument-adjuncts in the core (goals of motion and
change-of-place verbs) and argument-adjuncts in the periphery (benefac-
tives), a distinction that accounts for the different syntactic behaviour of
these kinds of adpositional phrases.
Following this line of argument and drawing on the RRG distinction
between semantic arguments and adjuncts, syntactic core and periphery,
and predicative and non-predicative prepositions, Ibáñez Cerda (2011) pro-
poses a principled system of eight logical types of adpositional phrases (APs).
Each type is defined in terms of a combination of features relating to the three
structural – semantic, syntactic and categorical – dimensions. The values for
each dimension can be defined as follows: (a) in consideration of the semantic
nature of the participant they code, APs can be [þ/ argument]; (b) in terms of
their syntactic status, they can be [þ/ core]; and (c) with respect to the
categorical status of their preposition, they can be [þ/ predicative]. The
combination of features gives us the following set of possibilities.
1. Oblique Core Argument
(þ) Argument (semantic level)
(þ) Core (syntactic level)
() Predicative preposition (intra-syntagmatic or categorical level)
2. Peripheral Adjunct
() Argument
() Core
(þ) Predicative preposition
3. Argument-Adjunct in the Core
(þ) Argument
(þ) Core
4. Argument in the Periphery
(þ) Argument
() Core
() Predicative preposition
5. Argument-Adjunct in the Periphery
(þ) Argument
() Core
6. Adjunct in the Core
() Argument
(þ) Core

7. Adjunct-Argument in the Core4

() Argument
(þ) Core
8. Not labelled
() Argument
() Core
Of the eight logical possibilities, only the last type, that is, the one defined
by the negative value of all the features, is ruled out by functional prin-
ciples: there is no way, it seems, in which a semantic adjunct, with no core
privileges, can be introduced by a non-predicative preposition. In the
following section, examples from English and Spanish of each of the other
seven types are provided.
10.5.1 Oblique Core Argument
(þ) Argument
(þ) Core
These are canonical oblique core arguments in RRG terms: semantic argu-
ments of the predicate in the nucleus of the clause, which are coded as APs
introduced by a non-predicative adposition. This is the case with the recipi-
ent argument of transfer verbs in English (36a) and of the goal of motion
and change-of-place verbs (36b–c), when they are introduced by canonical
prepositions, like a and en in Spanish, respectively, as these prepositions
are predictable from specific positions in the LS of those predicates.
(36) a. Tony gave the book to Peter.

b. Juan fue a la tienda.
John went to the store
‘John went to the store.’
c. Juan puso el libro en la caja.
John put the book in the box
‘John put the book in the box.’
10.5.2 Peripheral Adjuncts
() Argument
() Core
These are canonical peripheral clause participants. They are not semantically
required by the predicate in the nucleus of the clause, and thus they are

adjuncts. They cannot function as controllers, and hence they must be outside
of the core, that is, in the periphery. Finally, their preposition is predicative
and, as such, it licenses the participant as its argument. The most straightfor-
ward examples of this type of PP are the temporal and locative adjuncts that
function as settings of the state of affairs denoted by the predicate.
(37) a. John baked a cake after work.

b. John baked a cake in the kitchen.
10.5.3 Argument-Adjunct in the Core
(þ) Argument
(þ) Core
This specification of features implies that the participant is an argument,

that it is in the core, and that its preposition is predicative, and this is in fact
what differentiates this type of PP from the standard oblique core argument.
As seen in Section 10.4.3, in Spanish this type of adpositional phrase is
exemplified by the goal PPs of intransitive motion and change-of-place verbs,
but only when they are introduced by non-canonical prepositions. These add
a semantic specification to the referent of the verbal argument, and that is
why they are predicative, although both the argument and the preposition
variability are lexical features of the verbal item. As seen before, argument-
adjuncts in the core can control pivots in coordinate and subordinate
clauses, and hence they behave as core phrases.
(38) a. Juan fue hacia /hasta / adentro de la tienda.

John went towards /as far as inside of the store
‘John went towards /as far as / inside the store.’
b. Juan puso el libro sobre / dentro de / debajo de la caja.
John put the book on / inside of / under of the box
‘John put the book on / inside / under the box.’
10.5.4 Argument in the Periphery
(þ) Argument
() Controller
As expected from the iconic principle in the semantic–syntactic correlation,

semantic arguments of the verb are canonically in the core, that is, they
‘naturally’ have access to some syntactic privileges that set them apart from
canonical adjuncts, which in principle do not have those privileges and are
peripheral. Nevertheless, there are cases in which a semantic argument can
be deprived of all its possible syntactic privileges, and when this happens, it
belongs to the core-level periphery. A well-established case of this type is the

agent PP of the passive construction in English, in which the agent loses all
its syntactic privileges and, as a result, it cannot be projected as a core
argument. In fact, there are languages in which only the direct arguments
are in the core; that is, all oblique arguments seem not to have control or
pivot functions or any other type of syntactic privilege (Bickel 2003).
Other possible examples of this type are those APs which code participants
that are semantically required by the verb, for instance arguments, like en
francés in (39a) and con un gesto in (39b), but in an alternative construction
can be projected as subject or as direct object, as can be seen in (39c) and
(39d). This behaviour indicates that they are indeed arguments.
(39) a. Pedro habló con María (en francés).

Pedro talked with Mary (in French)
‘Pedro talked to María in French.’
b. Mauricio le declaró su amor a Tere (con un gesto).
Mauricio dat.3sg declared his love dat Tere (with a gesture)
‘Mauricio declared his love to Tere with a gesture.’
c. Pedro habla francés perfectamente.
Pedro speaks French perfectly
‘Pedro speaks perfect French.’
d. Su gesto lo declaró todo.
His gesture acc.3sg declared all
‘His gesture declared it all.’
Importantly, when these participants are coded as APs, they are clearly
optional and peripheral, so they have the status of arguments in
the periphery.
10.5.5 Argument-Adjunct in the Periphery
(þ) Argument
() Core
This type of AP is exemplified by the goal complement of Spanish motion verbs

which inherently imply a source, like salir ‘to go out’ and partir ‘to leave’.
(40) a. Lolai salió a la callej y i/j estaba sucia.

Lola went.out to the street and was dirty
‘Lola went out to the street and she/it was dirty.’
b. Marisai salió para la cabañaj y i/*j estaba sucia.
Marisa went.out for the cabin and was dirty
‘Marisa went out to the cabin and she was dirty.’
c. Ramóni partió al bosquej y i/j estaba sucio.
Ramón left to.art.def wood and was dirty
‘Ramón left for the wood and he/it was dirty.’
d. Toñoi partió para el bosquej y i /*j estaba sucio.
Toño left for the wood and was dirty
‘Toño left for the wood and he was dirty.’

As proposed in Ibáñez Cerda (2005), a goal complement may appear with

these verbs, although it is not inherently implied by them, because this
participant belongs to an enhanced movement frame, which permits those
verbs to be frequently coded with this argument. When the goal is intro-
duced by the canonical preposition a, it behaves as a core argument, as can
be seen from the fact that it can control a pivot in a coordinate clause as in
(40a, c). When the goal of salir and partir is coded with a non-canonical
preposition, like para in (40b, d), it cannot function as a controller, and this
signals that it is not in the core. One can say that in these examples, the goal
is an argument, but it is not syntactically focalized. Hence, in this case, the
goal is an argument-adjunct in the periphery.
10.5.6 Adjuncts in the Core
() Argument
(þ) Core
This AP type projects an adjunct, that is, a participant that is not required by
the predicate in the nucleus of the clause. Nevertheless, the (þ) core feature
indicates that, contrary to what is expected of adjuncts, this kind of AP
exhibits some important syntactic properties which give it core status. This
is the case with some Spanish manner, temporal and locative PPs, exempli-
fied in (41a–c).
(41) a. La mujer viste con elegancia.

The woman dresses with elegance
‘The woman dresses with elegance.’
b. María actuó en el momento adecuado.
María acted at the momento right
‘María acted at the right moment.’
c. El puente fue construido en el lado este de la ciudad.
the bridge was built in the side east of the city
‘The bridge was built in the east side of the city.’
d. *La mujer viste.
the woman dresses
e. *María actuó.
María acted
f. ??El puente fue construido.
the bridge was built
The PPs in (41a–c) are obligatory for the grammaticality of these clauses, as
can be seen from the comparison with their counterparts in (41d–f), which
are pragmatically odd and need some addition to become acceptable (e.g.
Finalmente, el puete fue construido ‘In the end the bridge was built’). Neverthe-
less, the participants projected through them are not semantically required
by their respective nucleus predicates, so they are adjuncts in the core.

10.5.7 Adjunct-Arguments in the Core
() Argument
(þ) Core
This kind of AP is found in one of the two dative constructions of Spanish. As

has been pointed out by Gutiérrez (1978) and Demonte (1994), among others,
in Spanish there are two types of dative construction: one is formed with PPs
which code recipient participants that are semantically required by the
predicate in the nucleus of the clause, and these are proper core arguments
(cf. 42a–b); the other dative construction exhibits PPs which code partici-
pants that are not inherent arguments of the verbs they appear with (see
42c–d).
(42) a. Susana (le) dio un regalo a María.

Susana (dat.3sg) gave a gift dat María
‘Susana gave a gift to María.’
b. Dulce (le) ofreció un vaso de vino a Julio.
Dulce (dat.3sg) offered a glass of wine dat Julio
‘Dulce offered a glass of wine to Julio.’
c. Mario le pintó la casa a Rosa.
Mario dat.3sg painted the house dat Rosa
‘Mario painted Rosa’s house (for her).’
d. Ramiro le trabajó dos años al Sr. Rodríguez.
Ramiro dat.3sg worked two years dat.def.art Mr. Rodríguez
‘Ramiro worked for Mr. Rodríguez during two years.’
e. *Mario pintó la casa a Rosa.
Mario painted the house dat Rosa
‘Mario painted Rosa’s house (for her).’
f. *Ramiro trabajó dos años al Sr. Rodríguez.
Ramiro worked two years dat.def.art Mr. Rodríguez
‘Ramiro worked for Mr. Rodríguez during two years.’
As can be seen from the contrast between (42c–d) and (42e–f), the PP of
(42c–d) obligatorily co-occurs with the clitic le, whereas this is not the case
with (42a–b). However, both types of dative PP can control pivots in non-
finite subordinate clauses.
(43) a. Juani dio un libro a Pedroj para _ j leer.

John gave a book dat Peter for read
‘John gave a book to Peter to read.’
b. Juan i le hizo un traje a Pedroj para _ j usar en la fiesta.
John dat.3sg made a suit dat Peter for wear at the party
‘John made Peter a suit to wear at the party.’
Therefore, we conclude that the second type of dative AP is an adjunct-

argument in the core.

10.6 Conclusion
This chapter introduced the standard RRG theory of adposition functions

and adposition assignment (Foley and Van Valin 1984; Van Valin 1991,
1993, 2005; Van Valin and LaPolla 1997; Jolly 1993). Starting from the
three-way classification of AP types that is standard in RRG, and capitaliz-
ing on the distinction between semantic argument and adjunct, syntactic
core and periphery, and the categorical distinction between predicative
and non-predicative preposition, we then discussed the principled system
of adpositional phrase types which was first proposed in Ibáñez
Cerda (2011).
References
Alarcos Llorach, Emilio. 1972. Estudios de gramática funcional del español.

Madrid: Gredos.
Bickel, Balthasar. 2003. Clause Linkage Typology. Lecture series delivered at the
2003 International Role and Reference Grammar Conference, UNESP, Sao
Jose do Rio Preto, Brazil.
Blake, Barry J. 2001. Case. Cambridge: Cambridge University Press.
Bresnan, Joan. 1982. The Mental Representation of Grammatical Relations. Cam-
Bresnan, Joan. 2001. Lexical-Functional Syntax, Malden, MA: Blackwell.
Cano Aguilar, Rafael. 1999. Los complementos de régimen verbal. In I.
Bosque and V. Demonte (eds.), Gramática descriptiva de la lengua española,
Vol. 2, 1807–1854. Madrid: Espasa.
Demonte, Violeta. 1994. La ditransitividad en español: Léxico y sintaxis. In V.
Demonte (ed.), Gramática del español, 431–470. Mexico: Colegio de México.
Dik, Simon C. 1989. The Theory of Functional Grammar, Part 1. Dordrecht: Foris.
Gruber, J. S. 1965. Studies in Lexical Relations. PhD dissertation, MIT. (Published
with revisions as Lexical Structures in Syntax and Semantics, Amsterdam:
North-Holland, 1976.)
Gutiérrez, Salvador. 1978. Sobre los dativos superfluos. Archivum XXVII–
XXVIII: 415–452.
Haiman, John. 1980. The iconicity of grammar: Isomorphism and motiv-
ation. Language 56: 515–540.
Hernández Alonso, César. 1990. En torno al suplemento. Anuario de Letras
XXVIII: 5–25.
Ibáñez Cerda, Sergio. 2005. Los verbos de movimiento intransitivos del español. Una
aproximación léxico-sintáctica. Mexico: ENAH-UNAM.

Ibáñez Cerda, Sergio. 2009. Prepositional phrases in RRG: A case of study

from Spanish. In L. Guerrero, S. Ibáñez and V. Belloro (eds.), Studies in Role
and Reference Grammar, 469–490. Mexico: UNAM.
Ibáñez Cerda, Sergio. 2011. PP types in RRG: A top-down approach to their
classification. In W. Nakamura (ed.), New Perspectives in Role and Reference
Grammar, 200–217. Newcastle upon Tyne : Cambridge Scholars Publishing.
Ibáñez Cerda, Sergio. 2021. Two-themes constructions and preposition
assignment in Spanish. In Robert D. Van Valin, Jr. (ed.), Challenges at the
Syntax–Semantics–Pragmatics Interface: a Role and Reference Grammar Perspec-
tive, 189–211. Newcastle upon Tyne: Cambridge Scholars Publishing.
Jolly, Julia. 1993. Preposition assignment in English. In Van Valin (ed.),
275–310.
Martínez García, Hortensia. 1986. El suplemento en español. Madrid: Gredos.
Nichols, Johanna. 1986. Head-marking and dependent-marking grammar.
Silverstein, Michael. 1981. Case marking and the nature of language. Austra-
lian Journal of Linguistics 1: 227–246.
Silverstein, Michael. 1993. Of nominatives and datives: Universal Grammar
from the bottom up. In Van Valin (ed.), 465–498.
Van Valin, Robert D., Jr. 1991. Another look at Icelandic case marking
and grammatical relations. Natural Language and Linguistic Theory 9:
145–194.
Van Valin, Robert D., Jr., and Randy J. LaPolla. 1997. Syntax: Structure, Meaning
Notes
1 RRG also posits a third parallel representation for information structure

(see Chapter 11), which is not relevant to our discussion.
2 Since Van Valin (2005: 44) the operator BECOME can be decomposed into
PROC & INGR, although for simplicity we will not adopt this
notation here.
3 The preposition con also appears in semantically determined LS contexts,
marking the non-undergoer theme of two-theme verbs, like conectar ‘to
connect’, reunir ‘to gather together’ and comparar ‘to compare’, among
many others (Ibáñez Cerda 2021).

4 Almost all of the proposed labels for the different PP types have been used
before in RRG studies, but they also arise directly from the combination of
features used in this work. For example, the label Argument-Adjunct in
the Core comes from the fact that the PP has the features (þargument),
(þcore) and (þpredicative preposition). This last feature gives the label the
‘adjunct’ part of the name, as it is a standard characteristic of adjuncts to
be formed with a predicative preposition. Nonetheless, the adjunct part is
attached to the ‘argument’ part of the name, because coding a participant
which is an argument is the main feature of this type of PP. I am now
introducing the label ‘Adjunct-Argument in the Core’ by means of the
same procedure: the combination of features is (argument), (þcore) and
(predicative preposition). That is, the main feature is that the partici-
pant coded in this type of PP is a semantic adjunct, so the first part of the
proposed name is ‘adjunct’. The ‘argument’ part of the name comes from
the fact that it is a standard characteristic of arguments to have non-
predicative prepositions. Finally, the ‘core’ part indicates that the PP has
some syntactic privileges.

11
The RRG Approach to
Information Structure
Delia Bentley
11.1 Introduction
In this chapter we introduce the theoretical constructs adopted by RRG in

the treatment of information structure, and we address the question of
where information structure fits in the architecture of a grammar.* Differ-
ent frameworks give different answers to this question. RRG is a parallel
architecture theory (Jackendoff 2002), which aims to describe and explain
cross-linguistic variation in the interaction of semantics, syntax and dis-
course pragmatics. To achieve this objective, RRG posits a direct linking
between semantics and syntax. Discourse constitutes an independent com-
ponent of grammar whose role is pervasive in the linking. A correlate of this
key aspect of the RRG approach to information structure is that, while some
syntactic positions are motivated in pragmatic terms, and it is recognized
that the left portion of the clause has pragmatic prominence in discourse-
configurational languages, there is no universal association of specific syn-
tactic positions or projections with specific discourse roles. Put differently,
the discourse properties and functions of individual syntactic constituents
are not contingent upon the placement in – or the displacement to –
particular syntactic positions. Rather, syntax and discourse are independent,
though crucially interfacing, components of grammar. In addition,
discourse-related meaning (for example, the distinction between the infor-
mation that has already been given and the new information that is pro-
vided with the utterance) is not only expressed syntactically, but also in
prosody, morphology and even in lexical choices, as is the case with the
selection of verbs with particular types of argument structure.
With respect to the constructs that are key in the treatment of infor-
mation structure, RRG follows Lambrecht (1994) in drawing a distinction
* I am grateful to Mitsuaki Shimojo, with whom I discussed several of the issues dealt with in the chapter.

The RRG Approach to Information Structure 457
between the notions of topic and focus, which are defined in terms of the
relations established between, and within, pragmatically structured propos-
itions and, on the other hand, the status of discourse referents in the minds
of the speech act participants. A key role in the RRG treatment of infor-
mation structure is also played by the distinction between presupposition
and assertion. The presupposition of an utterance is the information that is
shared by speaker and hearer prior to a sentence being uttered. The asser-
tion is the information that is known to the hearer as a result of the
sentence being uttered. Pragmatic relations and states will be dealt with
in Section 11.2. Then, in Section 11.3, we introduce the positions on the
layered structure of the clause that are motivated in terms of the encoding
of pragmatic relations.
The contrast between presupposition and assertion is reflected in focus
structure (Lambrecht 1994), that is, the conventional associations of focus
meanings with sentence forms. In accordance with the goal of typological
adequacy (Van Valin and LaPolla 1997: 8), these associations are assumed to
vary across languages. Focus structure is discussed in Section 11.4, where we
also introduce the distinction between the potential domain of focus in the
syntax of a given language and the actual domain of focus in an utterance of
that language. We then provide examples of cross-linguistic variation in
focus structure that depends on the variation in the potential domain of
focus as well as on the respective roles of prosody and syntax in the expres-
sion of discourse-related meaning (Section 11.5).
In Section 11.6 we discuss the pervasive role of discourse in linguistic
production and processing, illustrating it with reference to various steps
in the semantics–syntax linking. Lastly, we draw some conclusions
(Section 11.7).1
11.2 Pragmatic States and Pragmatic Relations
RRG adopts Lambrecht’s (1994: 49) distinction between non-relational and

relational constructs in information structure.2 Thus, on the one hand, it is
concerned with the status of the discourse referents in the minds of the
speech participants: whether they are already established or new from the
perspective of the hearer or both interlocutors, and, if they are new, whether
they can be individuated or, alternatively, related to previously introduced
referents. On the other hand, RRG takes information to be organized rela-
tionally and studies the relation between the presupposition, that is, the
information that can be taken for granted when an utterance is produced,
and the assertion, that is, the information which is known to the hearer as a
result of the utterance being produced. Although the two kinds of relational
and non-relational construct must be kept separate, and, indeed, research in
RRG suggests that they tend to be relevant to different steps in the linking
(Section 11.6), in due course we will highlight some alignment tendencies

458 DELIA BENTLEY
between the cognitive states of discourse referents and the packaging of

information in the utterance (Section 11.2.2). In the remainder of this
section, we discuss the non-relational and relational information structure
constructs in turn.
11.2.1 The Pragmatic Status of Discourse Referents

When a discourse referent is introduced for the first time, it is new from the
perspective of the speech participants or, simply, that of the hearer.3
Following Prince (1981), RRG differentiates between two types of new refer-
ent, called brand-new anchored and unanchored. The former type of refer-
ent, but not the latter, is explicitly related to referents that have already
been established, or can be identified, in discourse. Examples of referents
with these two kinds of cognitive status are provided by the noun phrases in
italics in (1a–b).
(1) a. I saw a student outside your door.

b. I saw a student from the linguistics department outside your door.
Not only is the referent of the italicized expression introduced as new in

both (1a) and (1b), as testified, in English, by the use of an indefinite article,
but in neither case can it be individuated by the speech participants. Thus,
the pragmatic state of this referent is unidentifiable in both cases, although
in (1b) the new referent is explicitly related to the known referent the
linguistics department. Contrastingly, in (1a), there is no such correlation.4
After they have been introduced, referents are normally treated as identi-
fiable, that is, as referents which can be individuated by the hearer. In
English, this may involve marking with a definiteness operator, like the in
(2a) and that in (2b), although we should note that cross-linguistically defin-
iteness does not necessarily signal identifiability and, vice versa, identifia-
bility does not necessarily correlate with definiteness.
(2) a. The student I saw outside your door asked me if I knew where you were.
b. That student asked me if I knew where you were.
Observe that the individuals referred to by the italicized expressions in

(2a) and (2b) are only identifiable insofar as this is not their first mention,
and thus a linguistic representation – or a file – has been created for them
in discourse (Lambrecht 1994: 77, see also Chafe 1976). Proper identifiabil-
ity involves the possibility for the hearer to individuate a single individual
or entity, or a single set of individuals or entities, that can be designated
with the linguistic expression chosen by the speaker. This case is illus-
trated in (3).
(3) The head of department is a semanticist.
Although many individuals can be designated with the expression head of

department, in this case the use of the definite article signals the speaker’s

expectation that the hearer will be able to single out the sole individual that
is being referred to with this expression in the given utterance.
The notion of frame or schema (Fillmore 1982; Lambrecht 1994: 90) can be
relevant to the treatment of a referent as an identifiable one. A frame or
schema is a system of associated concepts that are related in such a way that
to understand one of them one has to understand the whole system. These
systems of concepts justify the treatment of certain referents as identifiable,
because, when one of them is introduced, all the associated ones become
available. For example, if one is familiar with the concept of an academic
department, one will know that departments have heads. Once the notion of
department has been introduced, then the concept of the department’s head
becomes available.
Referents like those in the italicized expressions in (2a–b), which are to
some extent cognitively available with reference to the linguistic context or
(co-)text, are said to be textually accessible (recall that the intended context
of 2a–b is 1a–b). In some cases, these are referents that have been introduced
at an earlier stage in discourse and have not been mentioned for a while.
Other forms of cognitive availability are inferential or situational accessi-
bility, namely the property of an entity or an individual to be individuated
in the physical context, as is the case with (4b) or by means of some relation
with an element in the physical or linguistic context. The latter case is
illustrated in (5), where the referent of the noun phrase his head of department
may not be known to the hearer, but this referent is nonetheless accessible
by virtue of its relation to another referent (the latter is encoded by the
third-person pronoun his).
(4) a. «Which of these two classrooms do you teach in?»

b. «The large one.»
(5) His head of department is a semanticist.
Accessible referents are not in the current focus of attention prior to the
sentence being uttered (Van Valin and LaPolla 1997: 200). Identifiable refer-
ents that are in the hearer’s long-term memory, but neither in the short-term
memory nor in the current focus of consciousness, are called inactive,
whereas referents that are in the hearer’s current focus of consciousness are
activated or active.5 An active referent is encoded by them in (6).
(6) Remember John and Pete, the identical twins from high school? I saw them
in the lecture today.
Active referents can be deactivated and downgraded to a state of semi-

activity or mere accessibility due to the attentional and short-term memory
limitations of the interlocutors.
In Figure 11.1 we provide a graphic representation of the pragmatic states
introduced above.
The different strategies exhibited by languages for the coding of
referents, for example the marking with definiteness or indefiniteness

460 DELIA BENTLEY
Referential
identifiable unidentifiable
active accessible inactive anchored unanchored
textually situationally inferentially

Figure 11.1 The cognitive states of referents in discourse (Van Valin and LaPolla 1997: 201)
operators, correspond to different degrees of cognitive accessibility, reflect-

ing the speaker’s assumptions on the addressee’s knowledge and attention
at the time of the utterance, and ensuring that the least amount of
processing effort is made by the hearer (Van Valin and LaPolla 1997: 201).
An important background assumption is, thus, that the speaker makes
their utterance congruent with their assumption of the interlocutor’s
mental knowledge and disposition. Zero marking normally indicates
activation, while pronominal marking tends to correspond to activation
or accessibility, and so on. It should nonetheless be borne in mind that,
although such tendencies are recurrent cross-linguistically, languages
differ in their coding of the pragmatic states of discourse referents (Gundel
et al. 1993), and such cross-linguistic differences are within the domain of
investigation of RRG.
To give but one example, the importance of the notion of cognitive
accessibility is evidenced by Belloro’s (2004, 2015) RRG account of object
doubling in Argentinian Spanish dialects.
(7) Argentinian Spanish (Belloro 2015: 8–9)

a. Le gusta el cine a Juan.
him please.3sg the cinema to John
‘John likes cinema’
b. Lo vi a él.
him see.pst.1sg acc him
‘I saw him.’
Belloro points out that the object doubled by a clitic is not normally indefin-
ite in the dialects that she investigates. This fact would be puzzling, should
the key property of the doubled object be specificity, as was claimed in
earlier literature. Specifics can, in fact, be formally indefinite. While a
semantic analysis in terms of specificity is challenged by the evidence, an
information structure account in terms of identifiability would also seem to
be problematic on empirical adequacy grounds, since it is not the case that

all definite objects are doubled. The failure of some definite objects to be
doubled is captured with reference to the gradience of identifiability (see
Figure 11.1). Belloro (2015: 39) argues that the doubled objects encode
identifiable referents that are not active, but rather accessible. This claim
explains the facts exhaustively, predicting that the construction will not
exhibit indefinites, since indefinites encode unidentifiable referents or spe-
cifics which are unidentifiable from the perspective of the hearer. On the
other hand, it can host generics, in which case the discourse interlocutors
identify a class (Lambrecht 1994: 82, 88). An example of doubling of a generic
object is given here.
(8) Argentinian Spanish (Belloro 2015: 33)

[Nuestro voseo] los divierte mucho a los peruanos
our voseo them amuses much acc the Peruvians
‘Our voseo amuses Peruvians a lot.’
Object doubling in Argentinian Spanish thus turns out to be a strategy to

mark identifiable referents which may have temporarily fallen out of the
current focus of the hearer’s attention, and Belloro’s (2015) analysis of object
doubling in terms of the cognitive accessibility of the referent of the object
has explanatory power.
Other pragmatic states have been the object of research in RRG, for
example saliency, or persistence in discourse (Shimojo 2009), although due
to space limitations, we will not offer an account of such notions here. In the
next section, we turn to the pragmatic relations between units of infor-
mation that make the flow of information possible.
11.2.2 The Pragmatic Relations between Information Units

The construal and processing of information is not only analysed in
terms of the status of the discourse referents in the minds of the speech
participants, but also in terms of the relations between units of infor-
mation. An information unit (Van Valin 2005: 77) typically corresponds to
the information content of a single syntactic constituent, although it can
also refer to the content of a unit that is larger or smaller than a
syntactic constituent. Following Gundel (1988) and Lambrecht (1986,
1994, 2000), RRG defines topic as what the speaker wants to request
information about, or increase the addressee’s knowledge of, or get the
addressee to act with respect to (Van Valin 2005: 68). The definition of
topic is, therefore, inherently relational, in that it makes reference to the
information unit about which new information is being requested or
conveyed in the utterance. Thus, the information unit in italics in (9) is
not a topic by virtue of its information content, but rather it can be
considered to be a topic because the utterance increases the addressee’s
knowledge about it.
(9) The head of department has resigned.

462 DELIA BENTLEY
The topic is often presupposed, insofar as its status as a matter of interest

or concern in the utterance is part of the pragmatic presupposition. As was
mentioned in Section 11.1, the presupposition of an utterance is the set of
relevant propositions, or the information, which is shared by speaker and
hearer prior to the utterance.6 The topicality of the information unit in
italics in (9) is clear in the context of questions like ‘What did the head of
department do?’, since in this context, it is explicit that the head of department
is at issue.
Even in the absence of such a question, however, it can be claimed that the
unit the head of department is topical in (9). Indeed, in light of recent scholar-
ship (Cruschina 2012, 2015; Bentley et al. 2016, building on Reinhart 1981;
Frascarelli and Hinterhölzl 2007; Frascarelli 2017, among others) we propose
that a distinction ought to be drawn in RRG between aboutness and refer-
ential topics. Aboutness topics are referentially new because they are intro-
duced with the utterance, but they are topics in the relational sense of what
the utterance increases the addressee’s knowledge about. Referential topics,
on the other hand, are shared and referentially old, and, hence, they are
typically encoded by anaphoric expressions. In terms of this distinction, the
italicized information unit in (9) can be said to be an aboutness topic in an
out-of-the-blue context, since it is the subject matter of the sentence. By
contrast, the utterance in (10b) has a referential topic, whose status as the
current matter of interest is part of the presupposition.
(10) a. «What did the head of department do?»

b. «She resigned.»
Referential topics can – but need not – be detached in extra-clausal pos-

itions, with the consequent language- and construction-specific requirement
for a clause-internal resumptive pronoun (cf. 11a–b). There is no extra-
clausal detachment with aboutness topics (cf. 9).
(11) a. As for the head of department, she resigned.

b. As for the head of department, I saw her in the staff room.
The distinction between aboutness and referential topic highlights the

conceptual difference between what the speaker wants to increase the
addressee’s knowledge of, or to request information about, and the presup-
posed, shared, common ground, which the speaker relies upon, and which
both interlocutors know to be a current matter of interest. To be sure, a
referential topic can be both what the sentence is about and what speaker
and hearer know to be at issue. The examples in (10b) and (11a–b) are
illustrations of this, in that their referential topics are also what the sen-
tence increases the addressee’s knowledge of. According to an anonymous
reviewer, the fact that a referential topic can also be what a particular
utterance is about suggests that the notion of referential topic ought to be
subsumed within that of aboutness topic. While acknowledging that a
referential topic can have the aboutness function in an utterance, we

believe, however, that the two notions ought to be kept apart, in that only
referential topics are established in previous context. In fact, in textual
sequences, when a new aboutness topic is introduced, a previous aboutness
topic can take the role of a referential topic, which, however, will not play a
role in the topic-comment articulation of the utterance. The aboutness topic
of (12a), the proposal, turns into a referential topic in (12b), where the new
aboutness topic the teaching assistants is introduced.
(12) a. «The proposal was advanced unanimously by the students.»

b. «That’s right, but the teaching assistants were sceptical about it. In particular,
they felt that their role was misrepresented . . .»
This is an important reason for distinguishing referential and aboutness

topics. In Section 11.4, we shall point out some other advantages of this
distinction, which emerge in the analysis of focus structure. In this context,
we should stress that both kinds of topic are relational, insofar as a topic is
never identified in terms of its information content, but rather by virtue of
its relation to the information provided in the sentence or in the broader
text or discourse.
We should also mention in passing that since an utterance like (9) (The
head of department has resigned) presupposes knowledge about academic
departments having a head, it could be argued that the referent head of
department is part of the presupposition regardless of whether the sentence is
uttered in an out-of-the-blue context. We propose, however, that the kind of
knowledge mentioned here pertains to the notion of frame or schema (see
Section 11.2.1) and that this notion is relevant to the pragmatic states of
discourse referents rather than to their relations. Finally, the concept of
frame or schema is not to be confused with that of frame-setting topic,
which is an information unit that specifies the spatio-temporal coordinates
within which the eventuality described in the following utterance holds
true, as is the case with After the meeting in After the meeting, we all went for a
drink. In RRG, frame-setting topics are normally assumed to occupy the
extra-clausal pre-detached position (see Chapter 1, Section 1.6.1 and
Figure 1.34).7
The focus of an utterance is defined in RRG as what is asserted, in a
declarative sentence, and what is questioned, in an interrogative one (Van
Valin 2005: 69). Let us consider the following conversational exchange.
(13) a. «Who has resigned?»

b. «The head of department.»
The question in (13a) presupposes the proposition that someone has

resigned. The information conveyed by the reply in (13b), in turn, is not
the content of the information unit the head of department as such, but
rather the content of this unit in relation to the presupposed proposition
that someone has resigned. Only the information unit requested in the
question and that provided in the answer are focal in the respective

464 DELIA BENTLEY
Active Most acceptable

Accessible
Inactive
Brand-new anchored
Brand-new unanchored Least acceptable
Figure 11.2 The Topic Acceptability Scale (Van Valin and LaPolla 1997: 204)
utterances, but they are so with respect to the whole proposition of which
they are components.8 Accordingly, focus is defined with reference to the
pragmatic relation between two components of a proposition, the one
being presupposed, the other being the new contribution of the assertion,
or the contribution sought with a question. Thus, the notion of focus is
relational, on a par with that of topic. In the last analysis, the notions of
topic and focus are to be defined in terms of the speaker’s assessment of
the pragmatic relations between the components of a proposition in a
given discourse context. These notions will be discussed further in subse-
quent sections, after we have introduced other aspects of the RRG
approach to information structure.
Before we conclude this section, we return to the issue of the alignment of
cognitive states with pragmatic relations. Underlying the correlation
between these two types of pragmatic notion is the following principle:
the more accessible the topical information unit is, the less effort is required
to process an utterance. Accordingly, accessibility strongly aligns with top-
icality. On the basis of this principle, a Topic Acceptability Scale was
proposed by Lambrecht (1994: 165) and a slightly revised version was then
adopted in RRG (see Figure 11.2).
Recall that in the discussion of the cognitive states of discourse referents,
we pointed out that different strategies are adopted for the encoding of
different degrees of accessibility, thus ensuring that the least amount of
processing effort is made by the hearer. Zero marking tends to express
activation, while pronominal marking tends to correspond to activation or
accessibility, and so on. If we combine these coding strategies, ordered in
terms of the cognitive states that they express cross-linguistically, with the
Topic Acceptability Scale, we obtain the hierarchy shown in Figure 11.3,
which expresses the likelihood of coding of topic and focus by means of the
strategies that mark arguments in terms of their relative accessibility.
Figure 11.3 Coding of referents as topic and focus (Van Valin and LaPolla 1997: 205)

11.3 Information Structure and the Layered Structure

of the Clause
Before we discuss how presupposed and asserted information is packaged in

sentence structure, we briefly introduce the positions on the layered struc-
ture of the clause that tend to be associated with discourse functions. These
are the pre-detached position (PrDP), which hosts referential and frame-
setting topics, and the post-detached position (PoDP), which hosts after-
thoughts or referential topics.9 Contrastingly, the pre- and post-core slots
are filled by foci, although there can be restrictions on the kinds of foci that
they admit. In English, the pre-core slot hosts preverbal wh-words (14a) and
fronted constituents (14b).
(14) a. What did you read?

b. This book I have not read.
The pre-core slot can also, however, host topics in languages with a V2
constraint on word order (Diedrichsen 2008). As for the post-core slot, it
has been identified as the position of secondary foci that non-canonically
occur in postverbal position in Japanese, which is a verb-final language
(Shimojo 1995).
The pre-detached position can iterate, thus allowing the utterance to have
several topics, while the pre- and post-core slots cannot be repeated.
The core-internal positions and the peripheries of the various layers of the
clause can also host constituents with particular discourse roles. For
example, aboutness topics occur in the core-internal immediately pre-
nuclear position in SV languages. However, these positions are not pragmat-
ically motivated. In other words, they are not defined in terms of the
discourse roles that they host or in terms of the constructions in which they
are involved and which have pragmatic conditions on their occurrence. In
Figure 11.4, we indicate with grey shading, and lack thereof, the positions
that are pragmatically motivated and those that are not, respectively.
With respect to the issue of the relation between discourse and syntax,
RRG differs fundamentally from the syntactocentric approaches to linguis-
tic theory. On the one hand, it contrasts with Chomsky’s (1995) Minimalist
Program, which, at least in its original formulation, denied any status to
discourse in the syntactic computation, strictly associating discourse with
interface effects. On the other hand, it diverges from Cartography (Rizzi
1997, 2006), where movement operations targeting pragmatic positions start
PrDP Pre-Core Slot Core Nuc Core Post-Core Slot PoDP
topic focus (and topic) focus topic, afterthoughts

Figure 11.4 Pragmatically motivated (shaded) positions in the layered structure of the clause

466 DELIA BENTLEY
from a thematic position and land in a position associated with interpretive,

scope or discourse effects.
In RRG, arguments acquire their thematic roles by virtue of their pos-
ition in logical structure and not in syntax. Accordingly, the occurrence of
an information unit in a pragmatically motivated position does not
involve movement but rather the selection of a syntactic template, from
the syntactic inventory of the given language, which has the positions
needed to spell out the discourse functions of the given utterance. Import-
antly, the pragmatically motivated positions of the layered structure of the
clause are not claimed to be universal (Van Valin 2005: 8). Therefore, while
some languages have the possibility of expressing discourse roles in extra-
core syntax, by selecting a template with one or more of the positions
highlighted in grey in Figure 11.4, others solely spell out topic and focus
with morphological and prosodic clues within the core (see Sections 11.5,
11.6 and Chapter 12 of this volume for further discussion). No universal
one-to-one relation is assumed between discourse roles and functional
projections in RRG.
11.4 Focus Structure
The cross-linguistic variation in how presupposed and asserted information

is packaged in sentence structure is a concern of RRG. Following Lambrecht
(1994: 221–238), the conventional associations of information meanings
with sentence forms are referred to as focus structure. Three principal focus
structure types are distinguished; two of these involve the interplay of
presupposed and asserted information within the clause, whereas the third
one does not.
Predicate focus is considered to be the universally unmarked type of focus
structure. It expresses the type of proposition normally called categorical
judgement or statement, which was first identified by the philosophers Bren-
tano and Marty (see also Kuno 1972 and Sasse 1987, among others). This type
of proposition asserts or denies that all or some of the members of a set that is
defined by a category are members of another set defined by another category
(e.g. Some politicians are honest). In pragmatic terms, a predicate-focus structure
is construed as a comment about an information unit which is part of the
presupposition. The assertion, in turn, is an aboutness relation between the
focal predicate phrase and the presupposed topic. The examples in (15a–d),
where small caps indicate prosodic prominence within the asserted predicate,
are from Lambrecht (1994: 223).
(15) Question: «What happened to your car?»

Answer: a. «My car / it broke down.»
b. Italian
«(La mia macchina) si è rotta.»
the my car refl be3sg break.ptcp

c. French
«(Ma voiture) elle est en panne.»
my car it be3sg in breakdown
d. Japanese
«(Kuruma-wa) koshoo-shi-ta.»
car-top breakdown-do-pst
The topic of predicate-focus structures is typically treated as the subject of

the sentence. It has to be overt, albeit normally pronominal, in English and
French, but not in Italian or Japanese. In Japanese, a -wa-marked noun
phrase only occurs as the topic of a predicate focus structure if it is contrast-
ive (Shimojo 2010: 323 and Section 11.4).
In predicate-focus structures of languages without a grammaticalized
subject, for example Mandarin Chinese (LaPolla 1990, 1993, 1995 and see
Chapter 5), the topic is not treated as a subject.
(16) Mandarin (Van Valin and LaPolla 1997: 207, from Li and Thompson 1976)
Nèi xiē shù, shùshēn dà.
that few tree trunk big
‘Those trees, the trunks are big.’
Sentence structures which involve a topic vs. comment opposition but no

direct syntactic relation between the topic and the comment are also found
in languages with a grammatical relation subject, although these are usu-
ally pragmatically and syntactically more complex than the examples in
(15). An example is given here.
(17) Students, you have to remind them of the deadlines for coursework
submission every second week.
The topic of the utterance in (17) is a detached information unit in the pre-
detached position (see Figure 11.4), whereas the rest of the sentence is a
syntactic core (you have to remind them of the deadlines for coursework submission)
with an additional peripheral adverbial (every second week), both being within
the domain of the assertion.
As should be clear from the above, a defining feature of predicate focus
is the availability of the subject matter at issue from the presupposition.
Another defining feature of predicate focus can be explained with respect
to an important distinction made by RRG, that is the contrast between
the potential focus domain, which is the syntactic domain in the sen-
tence in which focus can occur in a given language, and the actual focus
domain, which is the syntactic constituent or constituents of a given
sentence that is or are in focus. In light of this distinction, predicate
focus is characterized by the limitation of the actual focus domain to the
syntactic core minus the subject-topic. This is shown in the speech act
projection in (15a), where the continuous line indicates the actual focus
domain, while the dotted line indicates the potential focus domain in
English.

468 DELIA BENTLEY
(15) a. [ Clause [ Core It [ Nuc broke down]]].
Speech Act Projection
Sentence focus differs from predicate focus insofar as the matter at issue
is not introduced previously in discourse, the whole proposition is asserted,
and the whole sentence is in the actual focus domain. Given that they do not
have a presupposed topic, sentence-focus structures occur out of the blue.
Lambrecht (1994: 223), the source of the following examples, also claims
that they reply to the question what happened?
(18) Question: «What happened?»

Answer: a. «My car broke down.»
b. Italian
«Mi si è rotta la macchina.»
poss.dat.1sg refl be3sg break.ptcp the car
c. French
«J’ai ma voiture qui est en panne.»
I have.1sg my car rel be.3sg in breakdown
d. Japanese
«Kuruma-ga koshoo-shi-ta.»
car-nom break.down-do-pst
(18) a. [ Clause [ Core My car [Nuc broke down]]].
The fact that the argument is in focus is reflected in the prosody of the
English construction in (18a) and in the morphosyntax of the constructions
in (18b–d). In both (18b) and (18c), the focal argument occurs postverbally.
Since French poses strict constraints on verb–subject order, the focal
argument is not encoded as a subject but rather as the undergoer of avoir
‘have’. Verb–subject order can also be found in English, although a locative
phrase or dummy there is required preverbally (There arrived three buses at the
station) and there are verb-class restrictions on this construction (Levin and
Rappaport Hovav 1995: 215–277). In Japanese, the focal argument is
marked with the nominative marker ga. The same focus meanings can
thus be expressed by different prosodic, morphological and syntactic struc-
tures in different languages.
Given that, in sentence focus, information is not structured as a contrast
between presupposed and non-presupposed information, this type of focus
structure expresses thetic statements, that is, unstructured statements
which describe a situation or an event, rather than making a statement
about a previously introduced individual or entity. Following Lambrecht

(1994), these sentences have traditionally been analysed in RRG as topicless

structures. However, scholarship on information structure increasingly sug-
gests that all utterances have a topic. Erteschik-Shir (1997) claims that a
predication is a function that maps a topic to a proposition, and that a truth-
value is assigned to the proposition in terms of the truth-value of the
predicate with respect to its topic. Utterances that do not have an overt
topic are argued by Erteschik-Shir (1997: 8, 26–29) to be assessed in terms of a
stage topic, which defines the spatial and/or temporal parameters of the
situation or event.
The adoption of the contrast between referential and aboutness topics
allows us to characterize properly the information structure of out-of-the-blue
statements, capturing the insight that all utterances have a topic. Thus, if
uttered out of the blue, the sentence in (9) (The head of department has resigned)
can be said to be a sentence-focus structure, which comprises an aboutness
topic (the head of department) that has not been introduced previously in
discourse. This sentence-focus structure differs from predicate focus, in that
its topic is not part of the presupposition or, put differently, it is not a
referential topic, and, thus, it is within the actual focus domain. In this
analysis the topic-comment articulation of the utterance does not correspond
to its partition into a background contrasted with the actual focus domain,
and we shall return to this point below. As for the stage topic of a seemingly
topicless sentence-focus utterance, in RRG it can simply be retrieved from
discourse and it need not be assigned a syntactic position.10
The third type of focus structure is called narrow focus (Van Valin 2005:
71). It has a single information unit – corresponding to a constituent or part
thereof – in the actual focus domain, whereas the remainder of the sentence
is part of the presupposition. The focal information unit can be an argu-
ment, an adjunct, or the verb. The examples in (19) are drawn from Lam-
brecht (1994: 223) (though the Italian examples in 19b have been slightly
modified).
(19) Question: «I heard your motorbike broke down?»

Answer: a. «My car broke down / It was my car that broke down.»
b. Italian
«Mi si è rotta la macchina.»
poss.dat.1sg refl be3sg break.ptcp the car
«È la macchina che mi si è rotta.»
be.3sg the car that poss.dat.1sg refl be3sg broken
«La macchina mi si è rotta.»
The car poss.dat.1sg refl be3sg broken
c. French
«C’est ma voiture qui est en panne.»
it be.3sg my car that be.3sg in breakdown
d. Japanese
«Kuruma-ga koshoo-shi-ta.»
car-nom break.down-do-pst

470 DELIA BENTLEY
(19) a. [ Clause [ Core My CAR [ Nuc broke down]]].
Lambrecht refers to this focus-structure type as argument focus to distin-

guish it clearly from predicate focus. Indeed, it could be argued that, when
the verb phrase is in focus, narrow focus cannot be distinguished from
predicate focus. This would be misleading, however, since, in predicate
focus, the domain of focus consists of the whole core minus the subject-
topic, and this definition subsumes structures with two coordinated or co-
subordinated cores (cf. 20). In narrow focus, instead, a single constituent or
part thereof is in focus (cf. 21).11
(20) a. «What did you do this morning?»

b. « [Core I [[ Nuc WORKED] [ Per IN THE LIBRARY]] [AND THEN [ Nuc LEFT ]]].»
(21) a. «When did you buy this house?»

b. « Actually, [ Core I [ Nuc RENT] it].»
At this juncture it should be noted that the distinction of actual focus

domain (AFD) and background (non-AFD) does not always correspond to the
topic–comment opposition, as was the case with (15a). Thus, in the reply to
the question in (22a), the topic is he, since the utterance is about the referent
of this pronoun and its antecedent John, whereas the verb bought is both part
of the background (the non-AFD) and of the comment on the topic John (see
Van Valin and Latrouite in Chapter 12 of this volume for further discussion
of this point).
(22) a. «What did John buy?»

b. «[ Topic He] [ Comment bought A CAR].»
Although bought a car is the comment on the topic he, only a car is in the
actual focus domain, since it is in narrow focus.

A distinction has been drawn between marked and unmarked narrow

focus (Van Valin and LaPolla 1997: 209–210). The latter type of narrow focus
occurs in an unmarked focus position in the clause, such positions being
clearly identifiable in many languages. The two English sentences below, for
instance, both illustrate unmarked narrow focus, in that the final position
in the core is the unmarked focus position in this language (cf. 23a), the sole
exception being wh-words, which are focal and occur by default in the pre-
core slot (cf. 23b).
(23) a. [Core He gave the book to Pat].

b. [Pre-Core S Who] [Core did he give the book to]?
However, since the potential focus domain extends to the whole clause, in
English, and focus is primarily marked prosodically, it can also fall in
marked positions, as is shown in (24).

b. [Core He gave the book to Pat].
c. [Core He gave the book to Pat].
When narrow focus falls in a marked position, it is contrastive or correct-

ive, in that it identifies a new selection out of a finite set of alternatives.
Thus, (24a–c) involve contrastive narrow focus in the respective contexts
listed in (25a–c).
(25) a. She gave the book to Pat.

b. He sold the book to Pat.
c. He gave the pen to Pat.
It should be noted, however, that narrow focus can be contrastive even when
it occurs in an unmarked position.
Contrastiveness as a relational information structure property that
selects from alternatives is orthogonal to the distinction between presuppos-
ition and assertion, in that the alternatives can be stated or predicted as part
of the presupposition. Thus, not only can foci be contrastive, but topics can,
too.
(26) a. «I heard Mary and Jane came yesterday.»

b. «Jane came.»
To the extent that its status as a matter of interest or concern in the

utterance has been introduced in (26a), the contrastive information unit in
(26b) (Jane) is presupposed. The new information introduced in the asser-
tion in (26b) is the relation between this information unit and the propos-
ition introduced in the previous utterance, as well as the notion that the
same relation does not hold true of the other referent introduced previ-
ously (Mary).
Interestingly, some languages allow the same marking of contrastive
topics and foci. This is the case with Japanese, where contrastive topics

472 DELIA BENTLEY
and foci can both be marked with -wa, although by default contrastive foci
are marked with -ga, and contrastive topics with -wa (see Shimojo 2009,
2010, 2011 for extensive discussion). To capture the topic–focus ambivalence
in topic marking in Japanese, Shimojo (2011) subsumes Erteschik-Shir’s
(1997, 2007) notion of subordinate f(ocus) structure within the RRG frame-
work. A subordinate f-structure is a complex f-structure involving embed-
ding. Contrastive information units can be topics or foci because there is a
contextually available set, which is topical, whereas the contrastive infor-
mation units that are selected from this set are foci. A structure with
contrastive -wa is thus represented as follows by Shimojo (2011: 275).
(27) [{xfoc, y}top]-watop [predicate]foc
In terms of sentence structure, Shimojo (2010, 2011) proposes that the -wa-
marked constituent figures outside the domain of focus in syntax, specific-
ally in the pre-detached position.12
11.5 The Interplay of Focus Structure and Syntax
Since typological variation is a key concern of RRG, attention has been paid
to the cross-linguistic variation in the interplay of prosody and syntax in the
expression of information structure (Van Valin 1999). In some languages, the
three types of focus structure discussed in Section 11.4 are primarily
encoded in syntax by means of word order, whereas in others they are
mainly encoded prosodically. In this second type of language, focus struc-
ture tends to be syntactically flexible, insofar as the potential focus domain
is not restricted to a particular portion of the clause.
Van Valin (1999) discusses this kind of typological variation in terms of
the relative rigidity or flexibility of focus structure and syntax across
languages. French, Toba Batak (a Western Austronesian language spoken
in Indonesia) and other languages have rigid syntax and rigid focus struc-
ture. This means that while the potential focus domain (see Section 11.4)
does not extend to the whole clause, focus cannot be freely marked by the
syntactic position of the focal unit either, since the syntactic position of
arguments is fixed to various degrees. In French, the actor of an active
clause occurs in the core-internal immediately pre-nuclear position. In the
spoken registers, it is usually omitted, or detached outside the clause,
while a coreferent clitic pronoun is spelled out preverbally. We shall not
dwell here on the status of this pronoun in morphosyntax, namely
whether it must be assumed to occur in a pre-nuclear position or, rather,
under the nuclear node (see Belloro 2004, 2015 for the position of core
clitic arguments in RRG). The point that is relevant in the current discus-
sion is that the actor of the active voice cannot normally occur in post-
nuclear position in French. Using terminology adopted in other frame-
works, French disallows free subject inversion (though see Lahousse 2011

and Lahousse and Lamiroy 2012 for further discussion). Focus, in turn, is
avoided in the core-internal immediately pre-nuclear position and is
marked post-nuclearly by default.13 The potential focus domain in the
French clause is thus as shown in (28).
(28) French
[PrDP Marie] [ Clause [ Core elle [ Nuc aime] les bonbons]] ]
Mary she love.3 SG the sweets

‘Mary loves sweets.’
Wh-words, which are focal, can occur preverbally, though in the pre-core
slot, rather than in a core-internal position. However, they can also occur in
situ, thus taking a post-nuclear position (cf. 29a–b). Otherwise, they are
clefted (cf. 30a).
(29) French
a. Vous avez vu qui?
you.2pl have.2pl seen who
‘Who have you seen?’
b. Vous allez o ù?
you.2pl go.2pl where
‘Where are you going?’
Given that the extent of the potential focus domain rules out focal
subjects in situ, while the syntax of French normally bans post-nuclear
subjects, the violation of these pragmatic and syntactic restrictions is
avoided by clefting (cf. 30b) or by using subjectless VS constructions (cf. 31).
(30) French
a. C’est qui qui a fait ça?
it be.3sg who who have.3sg done this
‘Who has done this?’
b. C’est Marie qui a fait ça.
it be.3sg Mary who have.3sg done this
‘It’s Mary who did this.’
(31) French
Il s’ est produit des problèmes.
it refl be.3sg produced some problems
‘There occurred some problems.’
In Toba Batak, which is a language with rigid VOS order and immediately
post-nuclear focus, the verb must carry a voice prefix when the undergoer is
topical and the actor is not (Van Valin 1999). This suffix indicates that
the undergoer has privileged syntactic argument status: the syntactic

474 DELIA BENTLEY
constraints on the position of O and S are satisfied, as is the constraint on

the position of focus. The postverbal actor receives a non-specific reading in
the postverbal position.
(32) Toba Batak (Van Valin 1999)

Di-jaha guru buku i.
pass-read teacher book the
‘The book is read by a teacher.’/‘A teacher is reading the book.’
The case of flexible syntax, which adapts to focus structure, is illustrated

by many Indo-European languages, for example Italian, and by Bantu, for
example Sesotho and Setswana (Demuth 1989, 1990). Whereas in both
groups of languages the potential focus domain is restricted to the nucleus
and the following constituents in the clause, as is the case with French
(cf. 28), in these languages the syntax allows VS order to mark a subject
as focal.
(33) Italian
[ Core [ Nuc Sono emersi] tanti problemi ].
be.3 PL emerged many problems
‘There emerged many problems.’
An important difference between Italian and Sesotho or Setswana is that,

in the two Bantu languages, there is an absolute constraint against pre-
nuclear foci, including wh-words. Thus, while Italian does have a pre-
nuclear focal domain, which is the pre-core slot (cf. 34), the two Bantu
languages do not.
(34) Italian
[PrCS Chi] [Core [Nuc ha fatto] questo]?
who have.3sg done this
‘Who did this?’
In Setswana, passivization or clefting of an actor is necessary to avoid a

breach of grammaticality (i.e. a subject wh-word in a pre-nuclear position).
(35) Setswana (Van Valin 1999)

*Mang o-pheh-ile lijo?
who sbj-cook-prf food
‘Who cooked the food?’

Lijo li-pheh-li-o-e ke mang.
food sbj-cook-prf-pass-mood by who
‘Who cooked the food?’ (Lit. ‘The food was cooked by who?’)


Ke mang ea o-f-ile-ng ntja?
cop who rel obj-give-prf-rel dog
‘Who gave you the dog?’ (Lit. ‘It is who that gave you the dog?’)
To return to Italian, not only does the pre-core slot host wh-words, but it also
admits contrastive narrow foci. This type of focus fronting is shown in (38b).
(38) Italian
a. «Cosa ha comprato Giorgio? La macchina?»
what have.3sg bought George the car
‘What did George buy? A car?’
b. «La bicicletta, Giorgio ha comprato, non la macchina.»
the bike George have.3sg bought neg the car
‘A bike George bought, not a car.’14
The fact that, in (38b), the focal argument co-occurs with the actor in pre-
nuclear position indicates that the position of the contrastive focus is core-
external, specifically, the pre-core slot (see Bentley 2008 for a fuller discussion
of this point). Observe that the preverbal actor must be assumed to occur core-
internally, specifically in the core-initial position. Indeed, RRG rules out the
assumption that a topic can follow the focus in the pre-core positions, that is,
those positions which, in other frameworks, are referred to as the left periph-
ery of the clause (Van Valin 2008: xx in disagreement with Rizzi 1997).
Languages with flexible syntax and flexible focus structure extend the
potential focus domain to the whole clause. Van Valin (1999) provides
relevant examples from Slavic, which are reported here.
(39) Russian (Van Valin 1999)

a. «Čto slučiloc’?»
what happened
‘What happened?’
b. «Mašina slomalac’.»
car broke.down
c. «Slomalac’ mašina.»
broke.down car
‘[My] car broke down’
The fact that both SV and VS order are allowed in sentence focus sets Russian
apart from the languages discussed previously, suggesting that preverbal
focus is not constrained in Russian in the same way as it is in Italian.
It should be noted, however, that the topic–focus order is adhered to in
declarative clauses with focus on the postverbal argument, while wh-words
occur in the pre-core slot (Van Valin 1999, based on Comrie 1979, 1984).
(40) Russian (Van Valin 1999)

a. «Kogo zaščiščajet Maksim-Ø?»
who.acc defends Maxim-nom
‘Who(m) does Maxim defend?’

476 DELIA BENTLEY
b. «Maksim-Ø zaščiščajet Viktor-a.»

Maxim-nom defends Viktor-acc
‘Maxim defends Viktor.’
Therefore, it can be hypothesized that, in Russian, core-internal pre-nuclear

focus is confined to sentence focus.
Further evidence that languages where the potential focus domain
extends to the whole clause may nonetheless exhibit restrictions on the
type of focus associated with particular sentence forms is provided by
Sicilian, which is cognate with Italian. Like Italian, Sicilian places contrast-
ive foci and wh-words in the pre-core slot (Bentley 2008, 2010).
(41) Sicilian (Bentley 2008)

a. «Chi ci accattau Pippinu a so niputi? A machina?»
what to.him buy.pst.3sg Joseph to his nephew the car
‘What did Joseph buy for this nephew? A car?’
b. «A bicicretta, Pippinu ci accattau.»
the bike Joseph to.him buy.pst.3sg
‘Joseph bought him a bike.’
Unlike Italian (cf. 42a), Sicilian also admits non-contrastive pre-nuclear

focus. In this case, the focal unit must be assumed to figure core internally,
as suggested by its scarce compatibility with another core-internal pre-
nuclear constituent (cf. 43b–c).
(42) Italian
a. «Che ha comprato Giuseppe a suo nipote?»
what has buy.ptcp Joseph to his nephew
‘What did Joseph buy for this nephew?’
b. «*La bicicletta ha comprato.»
the bike has buy.pst.3sg
‘He bought him a bike.’
(43) Sicilian (Bentley 2008)

a. «Chi ci accattau Pippinu a so niputi?»
what to.him buy.pst.3sg Joseph to his nephew
‘What did Joseph buy for this nephew?’
b. «A bicicretta (*?Pippinu) ci accattau.»
the bike Joseph to.him buy.pst.3sg
‘Joseph bought him a bike.’
c. «A bicicretta (*?a so niputi) ci accattau.»
the bike to his nephew dcl buy.pst.3sg
‘He bought a bike for his nephew.’
Data like (42)–(43) differentiate Sicilian from Italian, indicating that there
is no ban on core-internal pre-nuclear focus in Sicilian, and that the latter
language is more flexible than the former in terms of its focus structure.
Sicilian pre-nuclear focus, however, is not entirely free, but rather has
affective and emphatic connotations, which are absent from post-nuclear
focus (Sornicola 1983; Leone 1995; Cruschina 2006). In particular, preverbal

focus in (43b–c) suggests that the response being provided should be

expected by the hearer. In light of its special relevance effects, Sicilian
core-initial focus is not readily found in sentence focus.15
To sum up, while the extent of the potential focus domain allows languages
like Russian and Sicilian to express focus in any position in the clause, their
syntactic flexibility allows them to associate specific word orders with specific
types of focus structure or even to associate specific syntactic positions with
specific types of focus (as is the case with contrastive focus in the pre-core slot
and non-contrastive focus in the core-initial position in Sicilian). Therefore, in
these languages, the flexibility of focus structure amounts to the extent of the
potential focus domain alone and not to the lack of conventional associations
between focus meanings and sentence forms.
By contrast with Russian and Sicilian, English has flexible focus structure
but rigid syntax. Thus, while its syntax normally adheres to SVO order, the
prosody is tasked with the expression of focus structure. In (44) to (46), we
indicate the domain of focus in predicate focus, sentence focus and narrow
argument focus.
(44) a. «What happened to your car?»

b. «(My car)/It BROKE DOWN.»
Speech Act Projection (predicate focus)
(45) a. «What happened?»

b. «MY CAR BROKE DOWN.»
Speech Act Projection (sentence focus)
(46) a. «I hear your bike broke down.»

b. «MY CAR broke down.»
Speech Act Projection (narrow focus)
Narrow focus can fall in any position in the English clause (cf. 46b, 47a–d)
and is marked prosodically.

b. [Core He gave the book to Pat].
c. [Core He gave the book to Pat].
d. [Core He gave the book to Pat].
Adopting the Autosegmental-Metrical/Tones and Break Indices approach

to intonation (Pierrehumbert 1980), O’Connor (2008) has proposed a

478 DELIA BENTLEY
prosodic projection to represent the prosodic encoding of focus structure in

RRG. An important aspect of this proposal is that different languages are
said to avail themselves of different prosodic templates, comparable to the
syntactic templates which constitute the syntactic inventories of different
languages. Although O’Connor (2008) begins to establish the principles
which govern the interaction of the prosodic projection with the speech
act projection in English, to our knowledge his insights have not yet been
extended to the study of the prosodic encoding of information structure in
other languages.
Significantly, even languages with rigid syntax can deviate from their
default word order. For example, English allows fronting, in structures like
This book I did not read, and VS order, in there sentences and locative inversion
(There arrived three buses at the station) (see Section 11.4). In these cases, infor-
mation structure is not spelled out by prosody alone, but rather by the
choice of a particular constructional schema (Van Valin 2005: 131–135) in
semantics–syntax linking. It is to the role of discourse in the linking that we
now turn.
11.6 The Role of Discourse in the Linking
In RRG it is assumed that discourse operates at all stages of the semantics–

syntax linking, starting from the choice of a predicate for the clause (see
Chapter 12 for fuller discussion). Consider lexical alternations like the one
between actor and undergoer experiencer verbs, for example fear and
frighten. The availability of a topical stimulus as the subject matter at issue
may trigger the choice of frighten, which has a stimulus as its actor and
subject in the active voice (actor and subject align with topic cross-
linguistically).
(48) a. «Did you hear about the funding cuts?»

b. «Yes, it frightened everyone at the meeting.»
Conversely, the availability of an experiencer in the presupposition may

trigger the choice of fear, since this verb takes an experiencer as its actor
and subject in the active voice.
(49) a. «Do you like this resolution?»

b. «Actually, I fear its possible consequences.»
Once the logical structure of the predicate is retrieved from the lexicon,
its argument positions are filled with the core arguments, which are coded
in accordance with their pragmatic states and their roles as part of the
presupposition or the assertion (see Figures 11.2 and 11.3). Thus, in English,
the activated topical subject of a predicate-focus structure would most likely
be an unstressed pronoun, while in Italian and in the languages that allow
phonologically null subjects it would most likely not occur at all.

Phonologically null arguments can figure in logical structure with their

person and number features, when these features are spelled out in the
verb inflection.
(50) Italian
Parl-o.
speak-1sg
‘I speak.’
do′ (1sg act , [speak′ (1sg act )])
Alternatively, in the absence of an overt RP or of verb inflection, phonologic-

ally null arguments were previously assumed to be expressed by pro, which
is a silent variable that receives its value from discourse in the linking (Van
Valin 2005: 171–174). However, the current assumption in RRG is that there
is no silent pronoun in logical structure, consistently with the RRG ban on
null pronouns in syntax, and that the discourse antecedents of the unfilled
positions in logical structure can satisfy the valence requirements of the
verb. We return to this point at the end of the section.
The next step in the linking is macrorole assignment, which is not
affected by the discourse status of the arguments. Contrastingly, the latter
can play a key role in their morphosyntactic coding with case. We provide
here an example from Kaluli, an ergative language spoken in Papua New
Guinea (Van Valin 2015). In Kaluli, the predominant word order is OV, and,
as is the case with many languages, there is a strong association of actor
with topic. Ergative case is normally only marked on the topical actor when
the undergoer is animate. Otherwise, both actor and undergoer take
absolutive case (51). Pronominal arguments are marked by syntactic position
rather than case (52).16
(51) Kaluli (Van Valin 2015)

a. Abi-yò siabulu-wò mènigab.
Abi-abs sweet.potato-abs eat.3.tns
‘Abi is about to eat a sweet potato.’
b. Abi-yè Suela-yò sandab.
Abi-erg Suela-abs hit.3.tns
‘Abi hits Suela.’

a. E ne sandab.
3sg 1sg hit.3.tns
‘He/she hits me.’
b. Ne e sondòl.
1sg 3sg hit.1.tns
‘I hit him/her.’
However, when the actor is focal and the undergoer is not, the former is
marked by contrastive ergative morphology on pronominal actors and by
ergative case on non-pronominal ones, regardless of the animacy of the
undergoer.

480 DELIA BENTLEY

a. Nodo-wò niba diòl.
one.side- abs 1sg.contr take.1.tns
‘I (not you) take one side.’
b. Nodo-wò Suela-wè diab.
one.side- abs Suela-erg take.3.tns
‘SUELA takes one side.’
This complex split-ergative system is captured in RRG in terms of the effect

of discourse on case marking in the linking.
Voice alternations and the choice of privileged syntactic argument are
also affected by discourse. Thus, although accusative alignment normally
involves the selection of the actor as the subject, this default choice may be
overridden because of the role of actor and undergoer in the presupposition
and assertion. In particular, the passive, which maps the undergoer to
subject, may be chosen when this macrorole is topical and the actor is not.
An example is provided in (54), where the salience of an argument as the
matter at issue in the first utterance results in the marked subject assign-
ment in the second one.
(54) a. «The students’ proposal has many advantages.»

b. «Yes, I know. It will be discussed at tomorrow’s staff meeting.»
We pointed out above that passivization can also be a strategy to avoid

preverbal foci, as is the case with the Setswana data in (36).
The next step in the linking is the choice of a syntactic template from the
syntactic inventory. This is affected by discourse in several ways. To begin
with, passivization results in the failure of overt realization of the actor or
its occurrence outside the syntactic core. In such cases, the syntactic tem-
plate chosen for the clause does not have positions for all the arguments
within the core, but can host the actor in a periphery of the core (cf. 55b–c
vs. 55a).
(55) a. [Core The staff [Nuc will discuss] the proposal]

b. [Core The proposal [Nuc will be discussed]]
c. [Core The proposal [Nuc will be discussed]] [Per by the staff].
Secondly, we pointed out above that referential topics, afterthoughts, wh-

words and fronted foci occur in extra-clause or extra-core positions, in which
case the template selected for the sentence must include these positions.
(56) French (cf. 27)

a. [PrDP Marie] [Clause [Core elle [Nuc aime] les bonbons]]]
Mary she loves the sweets
‘Mary, she loves sweets.’
b. Italian (cf. 37a)
[[PrCS La bicicletta] [Core Giorgio [Nuc ha comprato,]]] [PoDP non la macchina.]
the bike George has bought neg the car
‘A bike George bought, not a car.’

Finally, in languages with flexible syntax, narrow focus and sentence focus
may involve marked word order patterns, for example VS, in languages with
default SV order (Section 11.5). Again, this is reflected in the choice of a
syntactic template for the clause, as is the case with the Russian example of
sentence focus previously given in (39c) and repeated below for convenience.
(57) Russian
[Core [Nuc Slomalac’] mašina].
broke.down car
‘[My] car broke down’
Prosody also plays a role in the linking, spelling out focus and providing
clues on the domains of presupposition and assertion in the processing of
the sentence. As was mentioned in Section 11.5, it has been proposed that
the role of prosody in information structure ought to be represented in a
prosodic projection (O’Connor 2008).
Although in this section we have discussed the role of discourse in the
linking from semantics to syntax, which pertains to language production,
we should note that discourse also plays a key role in the linking from
syntax to semantics, which pertains to language comprehension. By way of
conclusion of the section, we will briefly touch upon this direction in the
linking with reference to zero anaphora.
Zero anaphora is puzzling because it would seem to challenge the Com-
pleteness Constraint (Van Valin and LaPolla 1997: 325), which states that all
the arguments explicitly specified in the semantic representation of a sen-
tence must be realized syntactically in that sentence. In Mandarin Chinese
and other languages, a presupposed argument can be entirely silent (Van
Valin 2005: 174). Differently from the silent privileged syntactic argument of
many pro-drop languages (see the Italian example in (50)), the silent argu-
ment under discussion here has no pronominal or affixal expression on the
verb. Another puzzling case of zero anaphora is that of the Japanese verb-less
numeral quantifier construction, where an argument bears accusative case,
but there is no overt case assigner in clause structure. In this case, syntax
lacks a verb rather than an argument.
RRG captures these patterns of zero anaphora by linking the semantic
representation directly with discourse, drawing upon Kamp and Reyle’s
(1993) Discourse Representation Theory. Proper discussion of the application
of this theory to RRG is provided in Chapter 12, where Van Valin and
Latrouite propose an Extended Completeness Constraint, which states that
completeness can be satisfied by discourse representation structures in
combination with arguments that are expressed overtly in syntax. Here we
note that discourse representation keeps track of presupposed and asserted
information, which can be retrieved directly in the semantics of the clause,
without being expressed in syntax. Thus, the silent arguments of Mandarin
Chinese can be filled into the semantic representation in syntax–semantics
linking because they can be drawn from discourse representation, as can the

482 DELIA BENTLEY
predicator in the case of the Japanese verb-less numeral quantifier construc-

tion (for a detailed RRG treatment of this construction, we refer to Shimojo
2008). Importantly, this kind of direct linking is assumed to be bidirectional
and to apply to semantics to syntax linking as well. Thus, the argument
positions are linked directly with the discourse representation for zero
anaphora in semantics–syntax linking, and, in this kind of linking, the
predicator is linked directly to discourse in the Japanese verbless construc-
tion mentioned above.
11.7 Conclusion
Some of the most original and significant results of research in RRG suggest
that an important way in which languages differ from each other is in terms
of the role of discourse in the linking of semantics with syntax and syntax
with semantics. RRG offers an approach to information structure which is
flexible enough to capture this variation, while also being sufficiently con-
strained to make important generalizations regarding the expression of
pragmatic states and pragmatic relations, and their interface with prosody,
morphology and sentence structure across languages. In this chapter we
have introduced the constructs adopted in the RRG approach to information
structure and we have discussed the role that this framework gives to
information structure in the architecture of grammar.
References
Belloro, Valeria. 2004. A Role and Reference Grammar Account of Third Person Clitic
Clusters in Spanish. MA thesis, University at Buffalo (SUNY).
Belloro, Valeria. 2015. To the Right of the Verb: An Investigation of Clitic Doubling
and Right Dislocation in Three Spanish Dialects. Newcastle upon Tyne: Cam-
bridge Scholars Publishing.
Bentley, Delia. 2008. The interplay of focus structure and syntax. Evidence
from two sister languages. In Van Valin, Jr. (ed.), 263–284.
Bentley, Delia. 2010. Focus fronting in the layered structure of the clause. In
Nakamura (ed.), 4–28. https://rrg.caset.buffalo.edu/rrg/ProceedingsofRRG2009_
02.pdf.
from Italo-Romance. Linguistics 56: 1245–1301.
Bentley, Delia, Francesco Maria Ciconte, Silvio Cruschina and Michael Ram-
sammy. 2016. Micro-variation in information structure: Existential con-
structions in Italo-Romance. In M. M. Jocelyne Fernandez-Vest and Robert
D. Van Valin, Jr. (eds.), Information Structure and Spoken Language in a Cross-
Linguistic Perspective, 95–120. Amsterdam: John Benjamins.

Chafe, Wallace. 1976. Givenness, contrastiveness, definiteness, subjects,

topics, and point of view. In Charles Li (ed.), Subject and Topic, 27–55. New
York: Academic Press.
Chafe, Wallace. 1987. Cognitive constraints on information flow. In Russell
Tomlin (ed.), Coherence and Grounding in Discourse, 21–52. Amsterdam: John
Benjamins.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press.
Comrie, Bernard. 1979. Russian. In Timothy Shopen (ed.), Languages and Their
Status, 91–150. Cambridge, MA: Winthrop.
Comrie, Bernard. 1984. Russian. In W. S. Chisolm, Jr. (ed.), Interrogativity,
Cruschina, Silvio. 2006. Informational focus in Sicilian and the left periph-
ery. In Mara Frascarelli (ed.), Phases of Interpretation (Studies in Generative
Grammar 91), 363–385. Berlin: Mouton de Gruyter.
Cruschina, Silvio. 2012. Discourse-Related Features and Functional Projections.
New York: Oxford University Press.
Cruschina, Silvio. 2015. Focus Structure. In Delia Bentley, Francesco Maria
Ciconte and Silvio Cruschina (eds.), Existentials and Locatives in Romance
Dialects of Italy, 43–98. Oxford: Oxford University Press.
De Cia, Simone. 2019. The Syntactic-Pragmatic Interface in North-Eastern Italian
Dialects: Consequences for the Geometry of the Left Periphery. PhD dissertation,
University of Manchester.
Demuth, Katherine. 1989. Maturation and the acquisition of the Sesotho
passive. Language 65: 56–80.
Demuth, Katherine. 1990. Subject, topic and the Sesotho passive. Journal of
Child Language 17: 67–84.
Diedrichsen, Elke. 2008. Where is the precore slot? Mapping the layered
structure of the clause and German sentence typology. In Van Valin, Jr.
(ed.), 203–224.
Erteschik-Shir, Nomi. 1997. The Dynamics of Focus Structure. Cambridge: Cam-
Erteschik-Shir, Nomi. 2007. Information Structure: The Syntax-Discourse Interface.
Fillmore, Charles. 1982. Frame semantics. In The Linguistic Society of Korea
(eds.), Linguistics in the Morning Calm, 111–137. Seoul: Hanshin.
Frascarelli, Mara. 2017. Dislocations and framings. In Andreas Dufter and
Elisabeth Stark (eds.), Manual of Romance Morphosyntax and Syntax (Manuals
of Romance Linguistics 17), 472–501. Berlin: Mouton de Gruyter.
Frascarelli, Mara and Roland Hinterhölzl. 2007. Types of topics in German
and Italian. In Susanne Winkler and Kerstin Schwabe (eds.), On Information
Structure, Meaning and Form, 87–116. Amsterdam: John Benjamins.
Gundel, Jeanette K. 1988. Universals of topic-comment structure. In Michael
Hammond, Edith Moravcsik and Jessica Wirth (eds.), Studies in Syntactic
Typology, 209–239. Amsterdam: John Benjamins.

484 DELIA BENTLEY
Gundel, Jeanette K., Nancy Hedberg and Ron Zacharski. 1993. Cognitive
status and the form of referring expressions in discourse. Language 69(2):
274–307.
Jackendoff, Ray. 2002. The Architecture of the Language Faculty. Cambridge, MA:
MIT Press.
Kluwer.
Krifka, Manfred. 2007. Basic notions of information structure. In Caroline
Féry, Gisbert Fanselow and Manfred Krifka (eds.), The Notions of Information
Structure, 13–55. Potsdam: Universitätsverlag.
Kuno, Susumo. 1972. Functional sentence perspective: A case study from
Japanese and English. Linguistic Inquiry 3: 269–320.
Lahousse, Karen. 2011. Quand passent les cicognes: Le sujet nominal postverbal en
français moderne. Paris: Presses Universitaires de Vincennes.
Lahousse, Karen and Béatrice Lamiroy. 2012. Word order in French, Spanish
and Italian: A grammaticalization account. Folia Linguistica 46: 1–29.
Lambrecht, Knud. 1986. Topic, Focus and the Grammar of Spoken French. PhD
dissertation, University of California, Berkeley.
and the Mental Representation of Discourse Referents. Cambridge: Cambridge
University Press.
Lambrecht, Knud. 2000. When subjects behave like objects: An analysis of
the merging of S and O in sentence focus constructions across languages.
LaPolla, Randy J. 1990. Grammatical Relations in Chinese: Synchronic and Dia-
chronic Considerations. PhD dissertation, University of California, Berkeley.
LaPolla, Randy J. 1993. Arguments against ‘subject’ and ‘direct object’ as
viable concepts in Chinese. Bulletin of the Institute of History and Philology 63
(4): 759–813.
LaPolla, Randy J. 1995. Pragmatic relations and word order in Chinese. In
Pamela Downing and Michael Noonan (eds.), Word Order in Discourse,
Leone, Alfonso. 1995. Profilo di sintassi siciliana. Palermo: Centro di Studi
Filologici e Linguistici Siciliani.
Levin, Beth and Malka Rappaport Hovav. 1995. Unaccusativity: At the Syntax–
Lexical Semantics Interface. Cambridge, MA: MIT Press.
Li, Charles N. and Sandra Thompson. 1976. Subject and topic: A new typ-
ology of language. In Charles Li (ed.), Subject and Topic, 459–489. New York:
Academic Press.
Nakamura, Wataru (ed.). 2010. Proceedings of the 10th International Conference on
Role and Reference Grammar (RRG 2009). https://rrg.caset.buffalo.edu/rrg/Pro
ceedingsofRRG2009_02.pdf.
O’Connor, Rob. 2008. A prosodic projection for Role and Reference Gram-
mar. In Van Valin, Jr. (ed.), 227–244.

Pierrehumbert, J. 1980. The Phonology and Phonetics of English Intonation. PhD

Dissertation, MIT (published in 1988 by Indiana University Linguistics
Club).
Prince, Ellen F. 1981. Toward a taxonomy of given-new information. In Peter
Cole (ed.), Radical Pragmatics, 223–255. New York: Academic Press.
Prince, Ellen. 1992. The ZPG letter: Subjects, definiteness, and information-
status. In William C. Mann and Sandra A. Thompson (eds.), Discourse
Description: Diverse Linguistic Analyses of a Fund-Raising Text, 295–324. Amster-
Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence
topics. Philosophica 27: 53–94.
Rizzi, Luigi. 1997. The fine structure of the left periphery. In Liliane Haege-
man (ed.), Elements of Grammar, 281–337. Dordrecht: Kluwer.
Rizzi, Luigi. 2006. On the form of chains: Criterial positions and ECP effects.
In Lisa Lai-Shen Cheng and Norbert Corver (eds.), Wh-Movement. Moving On,
97–133. Cambridge, MA: MIT Press.
Sasse, Hans-Ürgen. 1987. The thetic-categorical distinction revisited. Linguis-
tics 25: 511–580.
Shimojo, Mitsuaki. 1995. Focus Structure and Morphosyntax in Japanese: Wa and
ga, and Word Order Flexibility. PhD dissertation, University at Buffalo
(SUNY).
numeral quantifier construction in Japanese. In Van Valin, Jr. (ed.),
285–304.
Shimojo, Mitsuaki. 2009. Focus structure and beyond: Discourse-pragmatics
in Role and Reference Grammar. In Lilián Guerrero, Sergio Ibáñez Cerda
and Valeria A. Belloro (eds.), Studies in Role and Reference Grammar, 111–141.
Mexico: Universidad Nacional Autónoma de México.
Shimojo, Mitsuaki. 2010. The left periphery and focus structure in Japanese.
In Wataru Nakamura (ed.), Proceedings of the 10th International Conference on
Role and Reference Grammar (RRG 2009), 315–235. www.acsu.buffalo.edu/
~rrgpage/rrg.html.
Shimojo, Mitsuaki. 2011. The left periphery and focus structure in Japanese. In
Wataru Nakamura (ed.), New Perspectives in Role and Reference Grammar,
266–293. Newcastle upon Tyne: Cambridge Scholars Publishing.
Sornicola, Rosanna. 1983. Relazioni d’ordine e segmentazione della frase in
italiano: Per una teoria della sintassi affettiva. In Paola Benincà, Michele
Cortelazzo, Aldo Prosdocimi, Laura Vanelli and Alberto Zamboni (eds.),
Scritti linguistici in onore di Giovan Battista Pellegrini, 561–577. Pisa: Pacini.
Stempel, Wolf-Dieter. 1981. L’amour elle appelle ça – L’amour tu connais
pas. In Christian Rohrer (ed.), Logos Semantikos. Studia linguistica in honorem
Eugenio Coseriu, Vol. 4: Grammatik, 351–367. Berlin: Mouton De Gruyter.
Tortora, Christina. 1997. The Syntax and Semantics of the Weak Locative. PhD
dissertation, University of Delaware.

486 DELIA BENTLEY
Van Valin, Robert D., Jr. 1999. A typology of the interaction of focus structure
and syntax. In Ekatarina Raxilina and Yakov G. Testelec (eds.), Typology and
Linguistic Theory: From Description to Explanation, 511–524. Moscow: Lan-
guages of Russian Culture.
Amazonian Languages. In M. M. Jocelyne Fernandez-Vest and Robert
D. Van Valin, Jr. (eds.), Information Structuring of Spoken Language from a
Cross-linguistic Perspective, 77–92. Berlin: De Gruyter Mouton.
Notes
1 The formal representation of discourse in RRG will not be touched upon,

as this will be dealt with in Chapter 12 by Latrouite and Van Valin. The
abbreviations used in the glosses of the examples are drawn from the
Leipzig Glossing Rules (www.eva.mpg.de/lingua/resources/glossing-rules
.php). See note 16 for an exception.
2 Lambrecht (1994: 49) discusses the distinction between ‘the pragmatic
states of the denotata of individual sentence constituents in the minds
of the speech participants’, and ‘the pragmatic relations established
between these denotata and the propositions in which they play the roles
of predicates and arguments’.
3 For brevity, we do not make explicit reference to the written language,
although it should be noted that a referent that is introduced for the first
time in a written text is new from the perspective of the reader.
4 As they are used here, the terms unidentifiable and identifiable corres-
pond with Prince’s (1992) concepts of hearer-new and hearer-old,
respectively.
5 The terms accessible, active and inactive are from Chafe (1987).
6 In Lambrecht’s (1994: 52) words, the presupposition is ‘the set of propos-
itions lexico-grammatically evoked in an utterance which the speaker
assumes the hearer already knows or is ready to take for granted at the
time the sentence is uttered’.
7 Some scholars compare frame setters to contrastive topics, insofar as
frame setters specify that the eventuality being described is only valid
within a particular domain or frame and may not be valid for other
domains (Krifka 2007). Assuming that this comparison stands typological
scrutiny, from the RRG perspective it does not necessarily follow that

frame setters and contrastive topics should take the same syntactic
position, although some evidence in support of this assumption is
reported in Frascarelli (2017: 486).
8 This notion of focus draws upon Lambrecht’s (1994: 213) claim that focus
is ‘the semantic component of a pragmatically structured proposition
whereby the assertion differs from the presupposition’.
9 The former names Left- and Right-Detached Position reflected a bias in
favour of left-to-right writing systems, differently from the new termin-
ology, which is based on order of occurrence in speech.
10 In the Gallo-Italian dialects of Northern Italy, the topic of sentence-focus
constructions with verbs of specific lexical classes can be expressed
overtly by a locative clitic in subject clitic position (Tortora 1997, and,
for an RRG account, Bentley 2018). The question arises of whether this
locative clitic should take a syntactic position. This question, however,
goes beyond the scope of the present chapter.
11 It should also be noted that the VP does not have any status in the
layered structure of the clause, because not all languages provide evi-
dence for it. Therefore, focus on the core minus the subject-topic cannot
be defined as focus on the VP constituent in RRG.
12 Further evidence that contrastive topics and foci can be marked in the
same way in individual languages has been found in Romance dialects
spoken in the north-east of Italy (De Cia 2019).
13 See Stempel (1981) for some deviations from this strong tendency
of French.
14 The definite marker in the RP can be said to be generic and it is not a
marker of specificity or uniqueness. Thus, this example could be trans-
lated as ‘now George is a bike owner, not a car owner’. One could also
find an indefinite RP in this example (Una bicicletta Giorgio ha comprato,
non una macchina ‘A car George has bought, not a car’), with the same
focus structure as discussed above.
15 Core-initial focus occurs in out-of-the blue ascriptive structures, where
the focus is on a non-verbal predicate (Sic. Stanca sugnu, lit. tired I-am).
16 The abbreviations contr and tns in the glosses of examples (51)–(53)
stand for ‘contrastive’ and ‘tense’, respectively.

12
Information Structure
and Argument Linking
Anja Latrouite and Robert D. Van Valin, Jr.
Abbreviations

A actor LS logical structure

AFD actual focus domain NUC nucleus
AV actor voice OCA oblique core argument
COMM comment PFD potential focus domain
DCA direct core argument PoCS post-core slot
DRS discourse representation PoDP post-detached position
structure
DRT Discourse Representation PrCS pre-core slot
Theory
CLM clause-linkage marker PrDP pre-detached position
GCG general common ground PRED predicator
IGC immediate common ground PSA privileged syntactic
argument
IF illocutionary force RLS realis
IS information structure RP reference phrase
IU information unit U undergoer
LS logical structure UV undergoer voice
12.1 Introduction
The RRG approach to information structure (IS) is laid out by Bentley

(Chapter 11), and the linking between syntax and semantics is explicated
by Van Valin (Chapter 1), Watters (Chapter 6), Guerrero (Chapter 14) and
París (Chapter 15). These discussions did not emphasize the role of IS in
linking, and it is to this topic that we turn in this chapter. We begin by

Information Structure and Argument Linking 489
presenting the representation of IS in the layered structure of the clause

(LSC; Section 12.2), and then we will show how context can be represented
using a version of Discourse Representation Theory (Kamp and Reyle 1993;
von Heusinger 1999) and the different focus types derived from these
representations (Section 12.3). The next section illustrates the importance
of IS for the analysis of grammatical phenomena (Section 12.4), and this
is followed by an analysis of conjunction reduction, which presented
problems for an RRG analysis in purely syntactic terms (Section 12.5).
Finally, the interaction of IS and the linking algorithm will be explored
(Section 12.6).
12.2 The Representation of Information Structure
Bentley introduced the distinction between the potential and actual focus
domains and the notion of information units, which are very important
parts of the RRG account of IS, along with representation of them in the LSC.
The LSC discussed so far has two projections, the constituent projection and
the operator projection. The theory posits two more, one for IS and one for
prosody (O’Connor 2008). They are treated as distinct projections, because
not all languages use prosody to signal IS (Van Valin 2016), even though
many, if not most, do.
The essential components of the focus structure projection as presented in
Van Valin (2005) are: (i) basic information units (IU), (ii) the potential
focus domain (PFD), and (iii) the actual focus domain (AFD). These are
illustrated in Figure 12.1.
The opposition between the PFD and the AFD is central to the RRG
analysis of IS. The PFD is a function of the grammar. In English the PFD
in simple clauses encompasses the entire clause, as can be readily seen in
SENTENCE
CLAUSE
CORE
RP NUC RP PP
PRED
Actual focus domain
V
Chris presented a child with some flowers
IU IU IU IU Basic information units

Potential focus domain
SPEECH ACT
Figure 12.1 Components of the focus structure projection (Van Valin 2005: 77)

490 ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.
Figure 12.2 (Unmarked) Narrow focus in English
Figures 12.1 and 12.2. In Italian, on the other hand, it excludes the pre-
nuclear core-initial position, as Bentley discusses in some detail (see also
Bentley 2008). The AFD is a function of the context in which the sentence
occurs; it is the focus of the utterance. It corresponds to the ‘focus domain’
of Lambrecht.
There is a very important IS distinction which is not represented here,
namely topic vs. comment. According to both Gundel (1988) and Lam-
brecht (1994), the topic referent is part of the pragmatic presupposition,
which is part of the immediate common ground (ICG) (Berio et al. 2017;
Krifka and Musan 2012). The ICG consists of the discourse context, which
includes the pragmatic presuppositions, plus the immediate social and
physical environment in which the speech event occurs. World knowledge
is contained in the general common ground (GCG). In Figure 12.1, there
is only a single IU outside of the AFD, namely, the RP Chris, which is
interpreted as the topic expression which refers to the topic referent. This
is not always the case, however, as it is possible for there to be more than
one IU outside the AFD, as illustrated in Figure 12.2.
The utterance represented in Figure 12.2 could be the answer to the
question What did John buy?, which establishes the referent of John and the
action of buying as part of the pragmatic presupposition in the ICG, and
the AFD is a new car. The part of the PFD which does not include the AFD is
termed the background.1 The individual referred to by the RP John is
presumably the topic, but this is not indicated in the representation. Balogh
(2021) proposes to augment the focus structure representation with nota-
tions indicating topic and comment. Following her proposal, the revised
representation would be as in Figure 12.3. Since it now codes both focus–
background and topic–comment, she labels it the ‘information structure
projection’.2

Figure 12.3 Revised IS-projection
Figure 12.4 The constituent, operator and information structure projections of a sentence
In Figure 12.3 he is indicated as the topic expression, and a new car is

presented as the (narrow) focus. The verb bought is part of both the comment
and the background.
It is possible to represent all three projections – constituent, operator and
information structure – in a single diagram, as shown in Figure 12.4. Each
projection represents a type of structural information which the grammar
may refer to, and the grammar can refer to the content of more than one
projection in the analysis of particular phenomena. For example, in Van
Valin (2005: 80–81) it is argued that one of the sources of VPs in languages
like English is the interaction of the information structure and constituent

projections: in predicate focus, as in Figure 12.1, the AFD demarcates a VP-

like grouping, which is a crucial part of conjunction reduction construc-
tions (topic-chaining), whereas in ‘VP’-ellipsis and -fronting, there is narrow
focus on the privileged syntactic argument (PSA) of the clause, leaving the
remainder as a VP-like grouping. In this case the VP-like grouping corres-
ponds to the background, whereas in conjunction reduction (topic-chaining)
it corresponds to the comment. (See (4) and (5) in Section 12.3 for more
discussion.) The IS-projection is not the only source of VPs, but it is an
important one.
12.3 The Representation of the Discourse Context
The focus constructions identified by Lambrecht (predicate focus, sentence

focus, and narrow focus) reflect the context in which the utterance occurs. It
is important, therefore, to have representations of the discourse context, in
order to justify the information structure of the utterance. It will also turn
out that the discourse representations will play a role in the linking in
certain constructions in some languages.
For the representation of the discourse context Van Valin (2005) and Shi-
mojo (2008) employ the version of Discourse Representation Theory (DRT)
(Kamp and Reyle 1993) proposed in von Heusinger (1999). In this system,
referring expressions and their coreference relations are represented in dis-
course representation structures (DRS), as in Figure 12.5.
The first sentence introduces the referring expression John, which is repre-
sented by the variable ‘v’, and the proposition expressed is ‘v stood up’.3 The
second sentence carries over the information in the first DRS (which is dark
grey rather than black due to its being less salient in the ICG), since that is
the context in which it is uttered, and it introduces two new referring
expressions, he and the door, which are assigned the variables ‘w’ and ‘x’,
John stood up. He opened the door. Mary kissed him.
v w, x y, z
John (v) John (v) John (v)

v stood up v stood up v stood up
he (w) door (x)
door (x) w=v
w=v w opened x
w opened x Mary (y)
him (z)
z=w
y kissed z
Figure 12.5 A simple example of DRSs

respectively. Since ‘w’ is a pronoun, its reference must be specified, and the
fact that it refers to ‘v’ is stated, and then the proposition ‘w opened x’ is
given. The final sentence is uttered in the context created by the first two
sentences, so the information from them is carried over, and the informa-
tion from the first DRS is light grey and that from the second is dark grey,
indicating their relative salience in the ICG. It introduces two more refer-
ring expressions, Mary (assigned ‘y’) and him (assigned ‘z’). Him being a
pronoun, its reference must be specified as ‘z ¼ w’, and the proposition in
it is ‘y kissed z’.
How are focus structures related to the DRSs? In Lambrecht’s approach the
focus is defined as the assertion minus the presupposition. Consider the
following question–answer pairs, in which the focus is in small caps.
(1) a. What did Bill do?

b. He bought a new car.
(2) a. What did Bill buy?

b. He bought a new car.
The (b) sentence is structurally the same in both pairs, but the focus is
different, due to the different presuppositions in the ICG established by
the questions. This can be represented as shown in Figures 12.6 and 12.7.
Figure 12.6 Derivation of predicate focus
Figure 12.7 Derivation of (unmarked) narrow focus
Central to the derivations is the introduction of two meta-variables, P

‘unspecified predicate’ and X ‘unspecified individual or entity’. P is

typically realized as do P ‘do something’ for an action, happen P ‘something

happened’ for an event, or be P ‘be something’ for certain kinds of states. In
Figure 12.6, the presupposition is a function of the question in (1a), and the
assertion is in (1b). Since the focus is the assertion minus the presuppos-
ition, it is bought a new car, with only the PSA he corresponding to an
element in the ICG. In (2), on the other hand, both Bill and buy are part of
the presupposition, leaving only a new car as the focus. Representing the
presupposition and the assertion separately amounts to breaking up
the non-initial DRSs in Figure 12.5 into the previous utterances(s) and the
current utterance.
What role can the DRSs play in the analysis of grammatical phenomena,
beyond explicating question–answer pairs like those in (1) and (2)? They turn
out to have a surprising number of applications, as will be shown in the
remainder of this chapter. To begin with, consider discourse-driven zero
anaphora in languages like Mandarin, Thai and Japanese, in light of RRG’s
rejection of phonologically null lexical elements in the syntax. The standard
way of dealing with them is to posit phonologically null pronouns (pros)
occupying argument positions in familiar tree structures, and the syntax can
manipulate the null pronouns exactly like the overt pronouns found in
familiar languages. Such an analysis was possible in earlier versions of RRG
(e.g. Van Valin and LaPolla 1997: 231–232) but not in contemporary RRG.4 The
key idea is that in languages like these three, discourse antecedents can
directly satisfy the valence requirements of verbs; in RRG terms, the Complete-
ness Constraint in these languages can be satisfied by overt RPs in the syntax
and discourse referents in DRSs. This obviates the need to posit null pronom-
inals, consistent with the RRG ban on null lexical elements in the syntax.
In order to illustrate how this works in practice, we will present the
analysis of the Japanese verb-less numeral classifier construction in Shimojo
(2008). The construction is given in (3b); (3a) is a simple clause with a
transitive verb.
(3) a. (Taro-ga) ringo-o ni-ko katta.

Taro-nom apple-acc 2-clf buy.pst
‘(Taro) bought two apples.’
b. (Oyatu-wa,) ringo-o ni-ko da.
snack-top apple-acc 2-clf be.prs
‘(As for snack,) [Taro got/will eat/etc.] two apples.’
[Literally: ‘As for snack, two apples [acc] is’]
In the first example, accusative case is assigned to the undergoer of the

transitive verb kau ‘buy’, which is the normal situation in Japanese. In the
second example, however, the RP ringo-o ni ko ‘two apples’ carries accusative
case (-o), but there is no transitive verb in the sentence; the only verbal
element is the copula, da, which expresses tense and which, importantly,
does not assign accusative case. The presence of tense indicates that this is a
well-formed clause, not a sentence fragment. The copula is a particular

feature of this construction, and without the numeral classifier the clause
with the copula and the accusative marking on the undergoer is ungram-
matical, unless it is an emphatic utterance to focalize ringo ‘apple’ (M.
Shimojo, pers. comm.).
The central question this construction poses is, ‘what licenses the accusa-
tive case on the RP in the absence of a transitive verb?’ A second question
concerns the interpretation of the actor in the sentence. It is a general
principle of Japanese that continuing, highly activated elements of the
ICG, especially topical actors (PSAs), are not normally expressed overtly.
Hence representing the activated constituents is crucial for explaining this
construction. Shimojo did an internet search for instances of this construc-
tion and looked at the first hundred examples. In eighty-nine of them the
actor and the verb were recoverable from the previous utterance; in eight of
them the interpretation was predictable from the preceding context (e.g. ‘I
just went to Starbucks. Two tall lattes-acc is,’ where the most likely mean-
ing is ‘bought’ or ‘drank’). The final three cases were online classified ads,
where the only verb that fits the context is ‘sell’. Thus, the interpretation of
the verb-less numeral classifier construction depends on either the immedi-
ately preceding utterance or contextually relevant sociocultural background
knowledge (the GCG). Shimojo’s RRG analysis based on representing context
via DRSs works for the eighty-nine discourse-context-dependent examples.
The preliminary step in the analysis in terms of linking from semantics to
syntax is to represent the activated information in the ICG; in this example,
it is ‘Taro got a snack’.5
The first two steps in the linking from semantics to syntax are to get the
logical structure (LS) of the predicate from the lexicon along with the
referring expressions that fill the argument positions, based on the message
to be expressed, first, and assign the semantic macroroles, second. This
yields Figure 12.8b.
The next step is to select the PSA and assign case. This yields Figure 12.8c.
This looks like a standard linking with a transitive verb in any accusative
language. But now the general principle that highly activated continuing
elements in the ICG are not normally expressed overtly in Japanese comes
into play. The next step involves the selection of the syntactic templates from
the syntactic inventory. The basic principle states that the number of argu-
ment slots in the core template corresponds to the number of specified
argument positions in the semantic representation, as a default, and it also
assumes a slot for the predicate in the nucleus. Based on these representa-
tions, it appears that the syntactic template should have two slots for core
arguments, but because the actor-PSA and the verb are part of the presuppos-
ition, they don’t figure in the determination of the core template. The only
part which is not part of the presupposition is ‘two apples’, and so this results
in the selection of a core template with only one argument slot (Figure 12.8d).
The final step is linking the accusative undergoer ‘two apples’ to the
argument position and inserting the copula in the nucleus to carry the

Figure 12.8a Representation of pragmatic presupposition for (3b)
Figure 12.8b Steps 1 and 2 of the semantics-to-syntax linking in (3b)
Figure 12.8c Step 3 of the semantics-to-syntax linking in (3b)
Figure 12.8d Step 4 of the semantics-to-syntax linking in (3b)
tense (not represented in these simplified semantic representations). The

actor RP Taro and the verb ‘GET’ do not appear overtly but are represented
solely in the DRS, as indicated by the curly brackets. This yields the assertion
in (3b) depicted in the ‘assertion DRS’ in Figure 12.8e.
The Completeness Constraint is satisfied by the presupposition DRS and the
syntactic representation. We may refer to the version of the constraint as found
in Japanese, Mandarin and Thai as the ‘Extended Completeness Constraint’.

Figure 12.8e Summary of linking from semantics to syntax in Japanese verb-less numeral quantifier
construction in (3b)
Figure 12.9 Summary of linking from syntax and discourse to semantics in Japanese verb-less
numeral quantifier construction
The presupposition DRS plays an equally important role in the linking

from syntax (and discourse) to semantics. The entire linking is summarized
in Figure 12.9.
The first step is the parser outputting a labelled syntactic tree. The
second step is the gleaning of the representation for indicators of the
function of the elements. In this example there is only one RP plus
the copula in the tree. The RP carries accusative case and so is determined
to be an undergoer. Accusative undergoers are associated with M-transitive
verbs, which leads to the expectation that there should be a nominative
actor and an M-transitive verb in the sentence, but there is neither a
nominative RP nor an M-transitive verb in the clause. However, in the
presupposition DRS there is an M-transitive verb with an actor argument
which would occur as a nominative RP, and so the conclusion is that the x-
argument in the DRS, Taro, is the actor which pairs with the overt under-
goer, ringo-o ni-ko. The next step is to go to the lexicon and retrieve the

logical structure of the verb identified in the previous step. But there is no
predicating verb in the sentence, and accordingly it is necessary to refer
again to the background DRS for it. The logical structure for ‘GET’ is called
up and its two arguments are assigned actor and undergoer macroroles.
The final step is the association of the arguments in the DRS and the syntax
with the arguments in the logical structure: Taro is the x-argument and the
actor of ‘GET’, and the a argument is the actor, hence Taro links to the a
argument; similarly, ringo-o ni-ko is an undergoer, and the b argument is
the undergoer in the logical structure of ‘GET’, therefore ringo-o ni-ko links
to the b argument. The resulting interpretation is given below the asser-
tion DRS. Here again the Extended Completeness Constraint is satisfied by
arguments from a DRS and the syntax.
This construction is particularly interesting, because it requires context
not only for the recovery of an argument, which is the usual situation with
discourse-driven zero anaphora, but also for the interpretation of the predi-
cate (verb). This is a phenomenon which is not limited to languages like
Japanese. English provides an example with ‘VP’-ellipsis across speakers, as
exemplified in (4b, b′).
(4) a. Kim is eating an ice cream cone, and Pat is, too.
a′. Kim is eating an ice cream cone, and so is Pat.
b. Speaker 1: Kim is eating an ice cream cone.
Speaker 2: Pat is, too.
b′. Speaker 2: So is Pat.
In Van Valin (2005: 231–235) the linking in (4a, a′) was presented in order to
show how RRG handles ‘VP’ phenomena despite not positing a VP node in
the constituent projection of the clause. Key to the analysis is the projection
of the semantic representation of the first clause onto the second clause in
the sentence, and this is possible because the two clauses are part of a single
sentence produced by a single speaker. This is not the case in (4b, b′),
however, as the two clauses are produced by two separate speakers in
separate utterances. Yet the interpretation is the same regardless of whether
the clauses are part of the same utterance or two utterances.
How can the interpretation be accounted for in the multi-speaker situ-
ation? The answer, as with the Japanese verb-less nominal classifier construc-
tion, involves DRSs. Instead of copying the semantic representation of the
first clause resulting from the syntax-to-semantics linking (with the PSA
replaced by a variable) onto the second clause, the result of the linking
constitutes the assertion made by the utterance, which is represented by a
DRS, and it becomes part of the ICG and functions as the pragmatic presup-
position for the second assertion. Because it is necessary for the construc-
tion, the DRS contains the tense–aspect values of the clause. The proposition
and operators contained in this DRS are projected as the semantic represen-
tation of the second utterance, and the single RP in the second utterance is

Figure 12.10 Linking from DRS in cross-speaker ‘VP’-ellipsis in English
linked as the PSA of the second clause. This can be summarized in

Figure 12.10. Thus, here again there is significant propositional information
recovered from context, which is represented by a DRS.
There is another interesting feature of this construction, which is not very
common in English, namely the option of having ‘RP AUX [, too]’ or ‘[So] AUX
RP’, as in both pairs of examples in (4). What influences the choice? A clue
can be found in ‘VP’-fronting, which shows a similar alternation.
(5) a. I expected to find someone mowing the yard, and mowing the yard was Bill.
a′. #I expected to find someone mowing the yard, and mowing the yard Bill was.
b. I expected to find Bill mowing the yard, and mowing the yard he was.
b′. #I expected to find Bill mowing the yard, and mowing the yard was he.
In the (5a) examples, the first clause does not introduce a specific referent for
the actor of mow, and when the referent is introduced, it is more felicitous to
have it after the finite auxiliary verb rather than before it; what we have here
is in effect ‘focal subject inversion’, rather like what is found in Italian and
discussed by Bentley in Chapter 11. When, on the other hand, the first clause
introduces a specific referent, as in the (5b) examples, the pronoun referring

back to it is strongly preferred in pre-auxiliary position. Thus, when the PSA is

focal, it occurs after the finite auxiliary, and when it is topical, it occurs before
it. This pattern seems applicable to ‘VP’-ellipsis in (4). When the participant
functioning as the PSA of the second clause or utterance has been mentioned
before, the ‘Pat is, too’ order is more appropriate, and the PSA, Pat, is a
contrastive topic in relation to the PSA of the first utterance or clause, in this
case, Kim. When, on the other hand, the PSA referent in the second clause or
utterance has not been mentioned before and is therefore new and focal, the
‘so is Pat’ order is more appropriate. Thus, these two constructions offer two of
the rare examples of IS-motivated ‘subject inversion’ in English, something
normally associated with Italian or Spanish.
12.4 Conjunction Reduction
The opposite of ‘VP’-ellipsis is conjunction reduction: in the former the PSA

is focal and is the only core argument in the linked clause, as in (4a),
whereas in conjunction reduction the PSA is topical and is missing from
the linked clause, for example, Bill is eating an ice cream cone and then will
drink a beer. As noted in Section 1.7.2 in Chapter 1, the PSAs of these two (or
more) clauses are coreferential, but there is no pronominal element in the
linked clauses that can be interpreted as coreferential with the PSA of the
first clause. It was suggested there that DRSs are the key to the analysis of
conjunction reduction, just as they are to the analysis of ‘VP’-ellipsis.
The crucial fact about this construction is that the controller–pivot rela-
tionship is restricted to the controller being the first RP in the core of the
first clause, and the pivots being realized by the absence of a pre-nuclear RP
in the core of the linked clauses. This means the controller is the PSA of the
initial clause, and the pivots are the PSA of each of the linked clauses. Hence,
in the constructional schema for conjunction reduction in Figure 12.11 a
crucial constraint is the restriction of coreference to the PSAs in
adjacent clauses.
The analysis of Mary bought a new car and drove it to Buffalo is given in
Figure 12.12.
PSA coreference is specified in the DRS, based on the constructional
schema, and the linking algorithm (given here in a reduced form) identifies
the PSAs in both clauses. The coreference statement ‘x ¼ v’ satisfies the
constraint, since both x and v are the PSAs of their respective clauses. The
PSA of the second clause is not lexically filled but is coreferential with
the PSA of the first clause, thereby avoiding a Completeness Constraint
violation. What happens when the PSAs are not coreferential? This is illus-
trated in Figure 12.13.
In this example, the PSA of clause 1, Mary (v), is not coreferential with the
PSA of clause 2, which is coreferential with a new car (y ¼ w), resulting in an

Figure 12.11 Constructional schema for English conjunction reduction
Figure 12.12 Analysis of conjunction reduction in English

Figure 12.13 Ungrammatical conjunction reduction in English
ungrammatical sentence. This was termed ‘grammatically constrained’ cor-

eference in Section 1.7.2 in Chapter 1, and it contrasts with ‘grammatically
unconstrained coreference’ as exemplified there by the data from Archi
(Kibrik 1979).
Thus, the RRG analysis of conjunction reduction of the kind found in
English, German, Icelandic and many other languages relies on the corefer-
ence statements in the DRSs and the grammatical constraints from the
constructional schema.
12.5 Information Structure and Linking between

Semantics and Syntax
In the remainder of this chapter we will explore the interaction of the

linking algorithm with IS, starting with the linking from semantics to
syntax, building on the introduction in Section 1.6. It will be shown that
IS can potentially interact with every step of the semantics-to-syntax linking.
Some of these interactions are found in all languages, for example the effect
of the activation status of a referent on the way the expression referring to it
is coded, while others, such as the effect of IS on PSA selection, range from
non-existent to crucial in different languages.

The major steps of the semantics-to-syntax linking algorithm are summar-

ized in (6).
(6) Linking algorithm: semantics to syntax (summary)

1. Construct the semantic representation of the sentence, based on the LS of
the predicator.
2. Determine the actor and undergoer assignments, following the Actor-
Undergoer Hierarchy.
3. Determine the morphosyntactic coding of the arguments
a. Select the PSA, based on the PSA Selection Hierarchy and associated
principles.
b. Assign the arguments the appropriate case markers and/or adpositions.
c. Assign the agreement marking to the main or auxiliary verb, as
appropriate.
4. Select the syntactic template(s) for the sentence following the template
selection principles.
5. Assign arguments and adjuncts to positions in the syntactic
12.5.1 Step 1
According to the model of a speaker proposed in Levelt (1989), the first step in
the production of an utterance is the formulation of an idea or thought to be
communicated, and the initial fully linguistic step is to convert it into a
semantic representation, which is step 1 in (6). This is primarily semantically
driven, but IS can have an influence in a number of ways. First, certain focus
constructions favour verbs of certain types. A good example is sentence focus
constructions, which have robust contextual constraints on their occurrence
(cf. Bentley, Chapter 11). Lambrecht (1987), along with Kuno (1972a), observed
that ‘the predicates most commonly permitted in [sentence focus] sentences
involve “presenting verbs”, i.e. intransitive verbs expressing appearance or
disappearance of some referent in the internal or external discourse setting,
or the beginning or end of some state involving the referent’(1987: 373).
A second example comes from languages without productive voice
systems. It was discussed in Chapter 5 on grammatical relations that IS
can influence PSA selection in languages with voice operations (see also
Section 12.5.3). It doesn’t follow, however, that IS is completely lacking in
influence in languages without voice. Languages may possess verbs which
present a state of affairs from different perspectives, as illustrated in (7).
(7) a. Bill gave a book to Sally.

a′. Sally received/took a book from Bill.
b. Covid-19 killed Bill’s cousin.
b′. Bill’s cousin died from Covid-19.
c. Sally led Bill through the park to the beach.
c′. Bill followed Sally through the park to the beach.

These pairs of sentences have different verbs which reflect different PSA
choices from among the participants in the state of affairs, and the selection
of a verb with its PSA reflects the salience of the referent of the PSA in the
discourse. In a context in which the referent of Bill is the topic and the
referent of Sally is newly mentioned, (7a, c′) would be more natural than (7a′,
c). Similarly, if the topic is the pandemic, then (7b) is appropriate, whereas if
Bill’s cousin is being talked about, then (7b′) is perhaps more natural. In a
language without voice, the primary way to vary the PSA in terms of the
different topicality statuses of the participants is through selecting a verb
which describes the same state of affairs as another verb but with a different
PSA, as in (7).6
A third, and very important, case of IS influence concerns the coding of
referring expressions. In (7a), for example, the referent of the giver of the
book is coded as a proper name, Bill. There are, however, a wide variety of
options available to a speaker, depending on the status of the referent in the
ICG: someone, a man, this guy, this/that man, the man, the man I met at the party
yesterday, the jerk/fool/idiot/genius, Bill, Mr Smith, he, that one, . . . All of these
different referring expressions can be used to denote the individual ‘Bill
Smith’, and the choice is a function of a number of factors, such as how
recently he has been referred to, how familiar the speaker believes the
interlocutor is with him. (See Bentley’s chapter on IS, Section 11.2, for a
detailed introduction to these distinctions.) In some languages, especially
head-marking languages, coding options include bound morphological pro-
nouns and anaphors, and clitics. The inventory of coding possibilities varies
from language to language, but in every language IS factors influence the
selection in actual utterances.
12.5.2 Step 2
The second step in (6) is macrorole assignment. In terms of the AUH, the actor
is the leftmost specified argument in the LS, and the default choice of under-
goer is the rightmost specified argument in the LS. This was shown in the
discussion of macrorole choice in Chapters 1 and 4. With respect to actor
selection, the only variation is whether the leftmost argument will be speci-
fied, that is, morphosyntactically instantiated in the clause, or left unspeci-
fied, as in an ‘agent-less’ passive, for example. In some cases this may be
motivated by the activation status of the referent of the leftmost argument, as
we saw in the Japanese example in (3b) in which highly activated PSA-
referents need not be explicitly mentioned again. In sentences involving
causal chains with two effectors, for example Sam opened the chest with the
skeleton key, the instigating animate effector can be left unspecified and the
intermediate inanimate effector, the instrument, can be selected as the actor,
yielding The skeleton key opened the chest.7 In both of these versions the actor is
the leftmost specified argument in the LS of the verb.

As pointed out in Chapters 1 and 4, undergoer selection is not absolute; in

some languages, for some verbs with three or more arguments, it is possible
to assign undergoer to an argument other than the rightmost. English is one
of the most permissive languages in this regard, as variable undergoer
selection is possible with most three-argument verbs. The most common
alternations are given in (8).
(8) a. Dative shift: Sam handed the letter to Alice vs. Sam handed Alice the letter.
b. Transfer alternation: Max presented the prize to Sally vs. Max presented Sally
with the prize.
c. Locative alternation: Mary sprinkled sugar on the cookies vs. Mary sprinkled the
cookies with sugar.
d. Removal alternation: Larry drained the water from the pool vs. Larry drained the
pool of its water.
In many cases, there is a semantic difference as a result of the different

undergoer choices, namely affectedness, as illustrated in (9).
(9) a. Dative shift:

Mary taught French to the students (but they didn’t learn a word of it).
Mary taught the students French (#but they didn’t learn a word of it).
b. Locative alternation:
The workmen loaded the boxes onto the truck. (all the boxes loaded, truck may
or may not be filled up; default interpretation)
The workmen loaded the truck with the boxes. (truck is full, all the boxes may or
may not be loaded; default interpretation)
In the dative shift examples the affectedness difference concerns whether

the students learned anything from Mary’s teaching: when the students is not
the undergoer, there is no implication one way or the other that they
learned anything, whereas when the students is the undergoer, it implies
that they did learn something, hence the oddity of the denial in the second
clause. The contrast in the locative alternation has been much discussed
over the past fifty years and needs no further comment.
When the alternation is semantically motivated, there is no direct role
for IS to play in it. In many cases, however, there is no semantic contrast, as
in (8a), and in such cases, IS can influence the choice of undergoer in
English, due to the difference in word order. In a language like German
or Russian, if the recipient is more topical than the theme, the order of the
dative and accusative RPs can simply be reversed. This is not an option that
is available to English speakers, as *Sam handed to Alice the letter or *Max
presented with the prize Sally is ungrammatical. Since the undergoer imme-
diately follows the verb in active voice cores, changing the argument
selected as undergoer changes the word order. Thus, the non-default
undergoer choice better reflects the topic-precedes-focus word order ten-
dency, as the following examples show. (‘>’ means ‘is preferred in this
context over’)

(10) a. Q: What happened with the letter?

A: Sam handed it to Alice > Sam handed Alice the letter/*it.
b. Q: Why is Alice upset?
A: Sam handed her the letter > Sam handed the letter to her.
(11) a. Q: What did Mary do with the sugar?

A: She sprinkled it on the cookies > She sprinkled the cookies with it.
b. Q: What did Mary do to the cookies?
A: She sprinkled them with sugar > She sprinkled sugar on them.
Each of the questions establishes the referent of either the recipient or the
theme as the topic in this limited context, and the more natural response to
it is the one in which the topic expression precedes the newly mentioned
referent. Thus, it is reasonable to conclude that there are cases in English
where IS can influence undergoer selection. It remains to be seen if there is
any evidence that this happens in other languages. German and Croatian,
for example, have limited variable undergoer selection, and it is doubtful
that it could have the same IS motivation as in English, since, as noted
earlier, simply reversing the order of dative and accusative RPs achieves the
same effect.
12.5.3 Step 3
The third step in the semantics-to-syntax linking is the assignment of mor-
phosyntactic functions and properties. The former concerns primarily PSA
selection and the latter agreement and case assignment. The influence of IS
on PSA selection in some languages was discussed in the chapters on
grammatical relations (5) and IS (11), and we will present a particularly
striking example of it. Ever since Schachter (1976), Tagalog has been seen
as a language in which IS plays a role in PSA selection. While this is often
discussed in terms of the definiteness or specificity of the undergoer being
crucial for PSA and voice selection (see LaPolla Chapter 5), there are add-
itional factors at work, as Latrouite (2011) shows. For example, causative
change-of-state predicates (e.g. tatay ‘kill’) strongly favour undergoer voice,
whereas activity predicates favour actor voice. Latrouite call this ‘event
structural prominence’, and it interacts with IS prominence in the deter-
mination of voice and PSA selection. A third factor is whether the actor or
undergoer argument is displaced to the beginning of the utterance.
Latrouite and Van Valin (2021), building on Schachter and Otanes (1972),
survey displacement constructions in Tagalog, and the two that are most
relevant to this discussion are ang-inversion and ay-inversion; the former is
illustrated in (12) with the verb sulat ‘write’.8
(12) a. S<um>ulat ng liham kay Lisa ang babae.

<av.rls>write gen letter dat Lisa nom woman
‘The woman wrote a letter to Lisa.’

b. S<in>ulat ng babae kay Lisa ang liham.

<uv.rls>write gen woman dat Lisa nom letter
‘A/The woman wrote the letter to Lisa.’
c. Ang babae ang s<um>ulat ng liham kay Lisa.
nom woman nom <av.rls>write gen letter dat Lisa
‘It was the woman who wrote a letter to Lisa.’
d. Ang liham ang s<in>ulat ng babae kay Lisa.
nom letter nom <uv.rls>write gen woman dat Lisa
‘It was the letter that a/the woman wrote to Lisa.’
(12a, b) represent the basic word order in a simple clause in Tagalog; (a)
illustrates actor voice, (b) undergoer voice. The nucleus is the first element
in the clause, and the default position for the ang-marked PSA is core-final.
The ang-inversion construction in (c) and (d) is usually translated by an
English it-cleft, which indicates that it should be interpreted as narrow focus
(see Latrouite 2021). In these examples the inverted RP is the nominative RP
congruent with the voice of the predicator in the nucleus, that is, in (c) the
verb is actor voice, and the displaced RP is the actor, while in (d) the verb is
undergoer voice, and the inverted RP is the undergoer. However, this correl-
ation does not always hold: Latrouite and Riester (2018) found that there
were cases where the actor was inverted but the verb was undergoer voice, as
exemplified in (13a). It is, however, not possible to invert the undergoer in
actor voice, as the ungrammaticality of (13b) shows.
(13) a. Ang babae ang s<in>ulat kay Lisa ang liham.

nom woman nom <uv.rls>write dat Lisa nom letter
‘It was the woman who wrote the letter to Lisa.’
b. *Ang liham ang s<um>ulat kay Lisa ang babae.
nom letter nom <av.rls>write dat Lisa nom woman
‘It was the letter that the woman wrote to Lisa.’
What explains this asymmetry? Latrouite and Riester argue that actor and
undergoer arguments have default IS values, with the actor being the
default topic and the undergoer being the default focus. With a topical
actor and an undergoer as part of the focus, IS does not require special
marking, and referentiality and verb semantics influence voice selection. If
the IS mapping is divergent, for example if the actor is topical (unmarked)
and the undergoer is topical (marked), we get undergoer voice to signal this
divergence. If the undergoer is a contrastive displaced topic (marked) and
the actor is topic (unmarked), we will get undergoer voice. If the actor is
focal (marked) and the undergoer is focal (not marked), then actor voice and
displacement (in the case of contrastive focus) tends to be used for
this divergence.
These default correlations are not surprising, since actors are the default
choice for PSA in the vast majority of languages (see Van Valin and LaPolla
1997, §6.5), and the default focus structure, predicate focus, has the PSA as

topic and the undergoer as focal. What is surprising is that IS can override
morphosyntactic constraints: focal actors are strongly preferred in displaced
(inverted) positions, as in (13a); note that the undergoer is the PSA and
therefore topical, which is the non-default IS function of undergoers. Hence
a displaced focal actor with a topical undergoer PSA and undergoer voice
reflects the marked IS-functions of the arguments. Why is (13b) impossible?
Two factors may be at work. First, the ang-marked lowest-ranking argument
of the verb is normally non-referential, which makes it a poor candidate for
displacement. Moreover, ang-marked arguments are always referential.
Hence the occurrence of ang-marking on the lowest-ranking argument is
incompatible with actor voice. In (12d), on the other hand, undergoer voice
signals that the lowest-ranking argument is referential, and therefore ang-
marking is appropriate. Thus one of the functions of the construction in
(12d) is to express a focal referential undergoer. This is a complex and
interesting example of the interplay of semantic macroroles and IS func-
tions with PSA and voice selection.
Agreement between the verbal complex in the nucleus and one or more
core arguments and case assignment shows an interesting asymmetry:
there appear to be many more examples of IS-influenced case assignment
than of IS-influenced agreement. The main examples of IS-influenced
finite verb or auxiliary agreement concern sentence focus constructions
in languages with restrictions on the PFD in simple clauses. It was pointed
out in Chapter 11 that in Italian the PFD in simple clauses excludes the
pre-nuclear position in the core; the AFD is restricted either to the pre-
core slot (PrCS) or to the nucleus plus post-nuclear constituents. Hence no
core-internal preverbal focus is permitted, and in the case of sentence
focus, the focal ‘subject’ must occur after the nucleus. It was also men-
tioned that in the southern Bantu languages Setswana and Sesotho
(Demuth 1989), the restriction is even stricter: no pre-nuclear focal mater-
ial is permitted at all. In sentence focus the focal argument must be post-
nuclear, and the predicate in the nucleus fails to show agreement. The
contrast between predicate focus and sentence focus is clear in (14), from
Demuth (1989).
(14) Setswana
a. Monna o-fihl-il-e.
man 3sg-arrive-pfv-mood
‘The/*a man arrived.’
b. Ho-filh-il-e monna.
loc-arrive-pfv-mood man
‘There arrived a man,’ or ‘A man arrived.’
Monna ‘man’ must be interpreted as topical, in accordance with its pre-

nuclear position in (14a), which is predicate focus. In the sentence-focus
construction in (14b), on the other hand, monna ‘man’ is focal and must
occur after the nucleus, and there is no agreement with it; the agreement

prefix in (a) is replaced by a locative prefix. Thus, in Setswana there is a

complex interaction among IS, word order and agreement, with the result
that in clauses with intransitive predicates there is agreement with the
single argument only when it topical and pre-nuclear. Italian, by contrast,
does show agreement with a post-nuclear argument in sentence-focus
clauses. Bentley (2018) investigates the interaction of word order (RP-V vs.
V-RP) and agreement in nine Italo-Romance varieties and finds variation in
whether the finite verb or auxiliary shows agreement in VS constructions. In
some, agreement is obligatory, like in Italian, in some it does not occur, like
in Setswana, and in some it is optional. What we have yet to find is an
example in which the PFD includes the entire clause, as in English, and
agreement with the PSA depends on whether the pre-nuclear RP is within
the AFD or not. This is represented in (15). (The AFD is represented by
underlining.)
(15) a. [CLAUSE [CORE RP [nucleus V/AUX[þAGR] . . . ]]]

b. [CLAUSE [CORE RP [nucleus V/AUX[AGR] . . . ]]]
In many languages there is agreement or cross-reference for more than

just the PSA with multi-argument verbs. In some languages, Lakhota
(Siouan), for example, undergoers of M-transitive verbs are coded on the
verb (in the third person primarily by a zero morpheme) regardless of
whether there is an overt undergoer RP or not and whether the argument ˇ
is topical or focal. In many Bantu languages, for instance Chichewa
(Bresnan and Mchombo 1987), Setswana (Demuth and Johnson 1989),
Kikuyu (Kihara 2017), on the other hand, the so-called ‘object marker’ is
in complementary distribution with an overt undergoer RP. When the
referent of the undergoer is focal, it must be expressed by a full RP
immediately following the verb; when it is not focal, it must be realized
by a bound pronominal on the verb, and an overt undergoer RP is not
possible in the same clause but if present must appear in a detached
position. Thus, there are clearly IS constraints on ‘object’ agreement or
cross-reference in these languages.
Finally, we turn to case assignment. A wonderful example of IS-influenced
case assignment from Kaluli (Papua New Guinea, Schieffelin 1985) was given
in Section 11.6. Korean exhibits two kinds of IS-influenced case marking,
which are called ‘case spreading’ and ‘case stacking’ (Park 1995; Han 1999).
Nominative case spreading is exemplified in (16) (from Park 1995), and
accusative case spreading and stacking are illustrated in (17) (from Han
1999).
(16) a. Thoyoil-ey kongcang-eyse pwul-i na-ass-ta. Basic form

Saturday-loc factory-loc fire-nom break.out-pst-decl
‘Fire broke out in the factory on Saturday.’
b. Thoyoil-i kongcang-i pwul-i na-ass-ta. Case spreading
Saturday-nom factory-nom fire-nom break.out-pst-decl
‘Fire broke out in the factory on Saturday.’

The contrast in (16) illustrates nominative case spreading with an intransi-

tive verb. The default form in (a) is an appropriate answer to the question
‘What happened at the factory on Saturday?’ The AFD encompasses pwuli
naassta ‘fire broke out’, with thoyoil-ey ‘on Saturday’ and kongcang-eyse ‘in/at
the factory’ being in the background within the PFD. Sentence (b), on the
other hand, would be an appropriate answer to ‘What happened?’, which
means that it is a sentence-focus construction in which the AFD includes the
entire clause. Thus it appears that the nominative case marks the AFD in this
construction. This correctly predicts that if the question were ‘What
happened on Saturday?’, then only kongchang ‘factory’ and pwul ‘fire’ would
be nominative, and that if it were ‘What happened at the factory?’, then only
thoyoil ‘Saturday and pwul would be nominative.
With transitive verbs it is the accusative case that is stacked or spread, and
the two forms have different semantic and pragmatic consequences.
(17) a. Chelswu-ka Yenghi-eykey kkot-ul cwu-ess-ta. Basic form

Chelswu-nom Yenghi-dat flower-acc give-pst-decl
‘Chelswu gave a flower to Yenghi.’
b. Chelswu-ka Yenghi-lul kkot-ul cwu-ess-ta. Case spreading
Chelswu-nom Yenghi-acc flower-acc give-pst-decl
‘Chelswu gave Yenghi a flower.’
c. Chelswu-ka Yenghi-eykey-lul kkot-ul cwu-ess-ta. Case stacking
Chelswu-nom Yenghi-dat-acc flower-acc give-pst-decl
‘Chelswu gave a flower to Yenghi.’
The basic form has default interpretations, both semantic and pragmatic.
The default interpretation is that the transfer of the flower to Yenghi was
completed, but this is not the only possible reading; it is also possible to say
without contradiction ‘but she did not receive it’. In terms of IS, the
unmarked focus position in Korean, as in other SOV languages, is the
immediate preverbal position (Kim 1988), and there are two ways to inter-
pret the AFD in predicate focus in (17a): it may not include Yenghi-eykey
‘Yenghi-dat’ (minimal AFD), or it may include it (extended AFD). In (17b)
the accusative case has replaced the dative on the recipient Yenghi; it has
‘spread’ from the undergoer to the non-macrorole recipient. When the
recipient is in the accusative case, it is impossible to add ‘but she did not
receive it’; it is a contradiction, because the accusative case signals that the
transfer was successfully completed. The accusative case canonically marks
the undergoer in Korean, and the undergoer codes the most affected partici-
pant in the event, as we saw in (9). Assigning the recipient accusative case
signals that its referent is affected by the action, which in the context of a
transfer event means that it got the item transferred.9
In (17c) the accusative case marker has been added to the dative-
marked recipient Yenghi-eykey, resulting in an RP with two case markers.
What does the accusative signal? It does not have the same

interpretation as the accusative in (b); it is possible to add ‘but she didn’t

receive it’ to (c) without contradiction. Rather, it signals that the dative
recipient is in the AFD; in other words, it marks the extended AFD. Thus
it has a clear IS-related function. What evidence is there that the stacked
accusative is a kind of focus marker? One piece of evidence comes from
wh-questions. In Korean the wh-word must appear in the preverbal focus
position, as in (18a); it is an instance of narrow focus. Stacking the
accusative case on the dative RP signals that this RP is in the AFD.
Combining case stacking on the recipient with an undergoer wh-question
generates a contradiction between the narrow focus of the wh-question
and the expanded AFD coded by the case stacking, and the result in (b) is
ungrammatical.
(18) a. Chelswu-ka Yenghi-eykey mwuet-ul cwu-ess-ni?

Chelswu-nom Yenghi-dat what-acc give-pst-q
‘What did Chelswu give to Yenghi?’
b. *Chelswu-ka Yenghi-eykey-lul mwuet-ul cwu-ess-ni?
Chelswu-nom Yenghi-dat-acc what-acc give-pst-q
‘What did Chelswu give to Yenghi?’
Here we have a clash of IS factors as the explanation for the ungrammatic-

ality of (18b), which supports the analysis of the stacked accusative as focus-
related. Han (1999) presents an extensive discussion of the full range of case-
substitution and case-stacking phenomena and formulates case-marking
rules to account for them; they supplement the basic RRG case-marking
rules for accusative languages introduced in Chapters 1 and 7.10
12.5.4 Step 4
The next step in the linking is selection of the syntactic template for the
utterance. The basic principles were introduced in Chapter 1, and they
made primary reference to the number of specified arguments in the LSs in
the semantic representation. The only reference to IS is indirect, mention-
ing displacement of arguments to the PrCS/PoCS (post-core slot) and the
occurrence of an argument-modulation voice construction, both of which
can be motivated by IS factors. The basic core syntactic templates, as
introduced in Chapter 1, are maximally unspecified and would not contain
any IS features. While this would be true for languages like English,
German, and all SOV and VSO languages, where the PFD includes the
whole clause, for SVO languages like French, Italian, Setswana and Seso-
tho, in which the PFD is constrained and excludes the core-initial, pre-
nuclear position, this could be indicated in the syntactic templates, since it
is a general constraint in the grammar of these languages. This is shown in
Figure 12.14, which represents the constraint that a pre-nuclear core argu-
ment is necessarily outside of the PFD. This constraint would be a feature
of all core templates in the language.

Figure 12.14 Restricted PFD in some SVO languages
Figure 12.15 Syntactic templates for English wh-question and locative-inversion constructions
There are syntactic templates that are specialized for specific IS properties,
and they would have the AFD specified on them, since it is a feature of the
construction. Examples from English include narrow focus on a wh-XP in the
PrCS and the locative inversion construction (e.g. Into the room walked a tall
stranger), which is always sentence focus. They are illustrated in Figure 12.15,
from Van Valin (2005).
In a language with a PrCS which always expresses narrow focus, regardless
of whether the XP in it is a wh-expression or a plain XP, then there would be
no [þWH] feature, and the extent of the PFD would be left unspecified.
Italian is interesting, in that it combines the restriction in Figure 12.14 with
the wh-template in Figure 12.15, yielding a two-part PFD which includes the
PrCS, on the one hand, and the nucleus and post-nuclear constituents, on
the other, but excludes the pre-nuclear core-initial position (Bentley 2008).
This can be seen in Figure 12.16, which is the structure of (38b) in Chapter 11.
It means ‘A bike George bought, not a car.’ (The grey shading indicates the
discontinuous PFD.)

Figure 12.16 Split PFD in Italian
The PoDP RP la macchina ‘a/the car’ is outside the scope of the illocutionary
force operator of the first clause, due to its being in a detached position. It is
therefore necessarily part of the presupposed ICG.
12.5.5 Step 5
The final step involves mapping the elements in the sentence, as developed in
steps 1–3, into the syntactic templates determined in step 4, to yield the
actual form of the sentence, morphophonological processes aside. A big part
of this is linearization of the constituents in the sentence, something that IS
has at least some influence on in every language and a great deal of influence
in many. Numerous examples of the influence on the linearization of con-
stituents, including displacement to the PrCS, for example, are discussed in
the earlier chapter on IS. No further examples need to be introduced here.
One very important area of the interaction of IS and linking concerns
extraction phenomena and constraints thereon, which are discussed by
Shimojo in Chapter 16. IS plays a central role in explaining constraints on
extraction phenomena, and this not only makes possible a more principled
and motivated account of cross-linguistic variation in this domain, but it
also forms the basis of an empirically testable account of the acquisition of
these constraints by children. See Van Valin (1994, 1998, 2008) and Van Valin
and LaPolla (1997, Epilog).
12.6 Linking from Syntax to Semantics
The ultimate purpose of IS is to facilitate the assimilation by the inter-

locutor(s) of the information which the speaker wishes to communicate

to them assuming the ICG. Hearers arrive at the IS of an utterance via a

complex set of considerations and calculations. In terms of the file card
metaphor for representing referents cognitively, the file card for the
referent of the topic expression is the one onto which the information
in the comment should be entered, with special attention paid to the
information in the AFD, which may require the creation of new file cards
for newly introduced referents. There are many aspects of the interpretive
process summarized in step 1 of the syntax-to-semantics linking algo-
rithm to which IS does not contribute. For example, as we have seen, IS
can affect PSA selection in some instances, and this in turn affects voice
selection, if the language has voice. As we have suggested, this is the
speaker’s perspective on the linking. From the hearer’s perspective, on
the other hand, the IS influence on PSA selection will normally have a
positive effect on information assimilation by keeping PSAhood aligned
with topicality, in that the default for the topic expression is that it
functions as the PSA and in most language occurs clause-initially. Devi-
ations from the expected correlation of PSAhood, position and topicality
are signalled through special prosody or special morphosyntactic con-
structions such as displacement of a contrastive RP to the PrCS in some
languages, use of a morphological contrastive focus marker in others. In
Section 12.5.3 it was shown that in Tagalog there are default discourse
statuses for actor and undergoer arguments with a multi-argument verb,
namely, topical for actors and focal for undergoers. Deviations from this
are signalled overtly: focal actors are obligatorily preposed, while topical
undergoers must be the PSA with undergoer voice on the verb. So, from a
hearer’s perspective, the interplay of position, voice and IS status gives
important clues to the interpretation of arguments.
Three important phenomena at the syntax–semantics interface which
crucially involve IS for their interpretation are determining the scope of
negation, determining the scope of quantifiers, especially in sentences
with multiple quantifiers, and the resolution of intrasentential pronomi-
nalization. Negation is the only operator that can function on all three
layers of clause structure, and one might reasonably expect that lan-
guages would have three distinct negative morphemes, one for nuclear
negation, one for core negation, and one for clausal negation, in order to
avoid scope ambiguities. One might, therefore, be disappointed to find
out that this rational pattern rarely occurs in human languages, and far
more common is the English pattern with a single primary negative
morpheme, not in English, which occurs in a restricted range of pos-
itions, and which can be interpreted as having nuclear, core or clausal
scope. IS provides the means to disambiguate the scope of negation. As
argued in §5.5 in Van Valin and LaPolla (1997), the scope of negation in a
clause is the AFD. This is illustrated in (19) with the sentence Lucy didn’t
buy a new car.

(19) a. Lucy didn’t buy a new car. Sally did. Marked narrow focus
b. Lucy didn’t buy a new car. She leased one. Marked narrow focus
c. Lucy didn’t buy a new car. She bought a used Unmarked narrow focus
pickup.
d. Lucy didn’t buy a new car. She leased a new pickup. Predicate focus
e. Lucy didn’t buy a new car. Nothing happened. Sentence focus
Each of the possible interpretations of the scope of not corresponds to one of

the focus structures proposed by Lambrecht and used in RRG, and is sig-
nalled primarily by intonation.
The interpretation of quantifier scope, especially in sentences with mul-
tiple quantifiers such as Every girl kissed a boy, has been the topic of much
discussion in the generative era, and explanations have been offered in
purely syntactic terms, requiring the postulation of covert movement, as
well as in formal semantic terms. However, Sgall et al. (1986), Kuno (1991),
Van Valin and LaPolla (1997: §5.5), Kuno et al. (1999) and Van Valin (2005: §3.6)
have argued that IS plays a crucial role in the interpretation of these sen-
tences. The basic principle is: everything else being equal, the more topical
quantifier has wider scope than the less topical quantifier. In a nutshell,
‘topic Q > focal Q’. For a sentence like Every girl kissed a boy, the default
(unmarked) interpretation is every > a, that is, every girl kissed a different
boy, which correlates with predicate focus, the default focus structure, in
which every girl is the topic expression and a boy is in the AFD. The non-
default (marked) interpretation, a > every, that is, there is a boy such that
every girl kissed him, is possible if the focus structure is marked narrow
focus on the PSA, leaving a boy to be the topic expression. This account makes
a number of significant and correct predictions about multiple quantifier
sentences cross-linguistically, and data from Italian, Japanese, Mandarin and
Toba Batak (Austronesian) are analysed in Van Valin (2005). An interesting
interaction of focus structure markedness and linking markedness has not
previously been discussed, to our knowledge. Consider Table 12.1 regarding
(i) Every girl kissed a boy and (ii) A boy was kissed by every girl.
The default (unmarked) reading for these sentences has the quantifier in
the PSA RP interpreted as having wide scope, as in (a) and (d). To get the other
reading requires a marked focus structure, but it appears that these readings
are not equally easy to get. Native speakers seem to have a more difficult time
Table 12.1 Interaction of focus structure markedness with linking markedness
Sentence Interpretation marked voice marked focus structure
(i) a. every > a marked marked

b. a > every marked þmarked
(ii) c. every > a þmarked þmarked
d. a > every þmarked marked

getting the reading in (c) than in (b). Why should this be the case?
A comparison of (a) and (c) suggests the following: (c) is difficult because it
combines the marked (non-default) voice (passive) with marked narrow focus
on the PSA, a doubly marked sentence, which yields the same interpretation
as the maximally unmarked form, (a). In processing terms, the hearer must
process two non-basic patterns to arrive at the same interpretation that the
two basic patterns would yield, and it is difficult to justify the processing load.
Hence the preference for (a) over (b) with (i) is weaker than the preference for
(d) over (c) with (ii). This is an interesting interaction of IS with syntax, namely
PSA and voice selection, with semantic consequences.
The final area is intrasentential pronominalization. Following the pion-
eering work on IS and pronoun resolution by Kuno (1972a, b, 1975), Bick-
erton (1975) and Bolinger (1979), Van Valin (1990) and Van Valin and LaPolla
(1997: §5.6) proposed an analysis of sentence-internal pronominalization
based on the following principle.
(20) Principle governing intrasentential pronominalization (Van Valin and LaPolla

1997: 227):
Coreference is possible between a lexical [R]P and a pronoun within the same
sentence if and only if
a. the lexical [R]P is outside of the actual focus domain, and
b. if the pronoun is in a syntactic argument position and precedes the lexical
[R]P, there is a clause boundary between the pronoun and the lexical [R]P.
Both of these conditions involve IS. The first one is illustrated in (21).
(21) a. Sami saw his i/j sister.

b. Sam i saw his*i/j sister.
In (21a) Sam is the topic expression, and his sister is in the AFD, and accord-
ingly, Sam can be interpreted as being coreferential with his. In (b), on the
other hand, Sam is in the AFD and therefore cannot be interpreted as
coreferential with his. The second principle in many instances involves a
contrast between the pre-detached position (PrDP) and the PrCS, and the
phrases in them differ in terms of their IS function.
(22) a. In Billi’s neighbourhood, hei is a big deal. PrDP

a′. In hisi neighbourhood, Billi is a big deal. PrDP
b. *In Billi’s living room hei put the new plant. PrCS
b′. In hisi living room Billi put the new plant. PrCS
In the example in (22a, a′) there is a PP containing a possessive RP in the

PrDP, and in (22a) the possessor is a lexical RP and is coreferential with the
pronominal PSA of the following clause, whereas in (a′) the possessor in the
PrDP is pronominal and the PSA of the following clause is a lexical RP. If
both are predicate focus, then coreference is possible in both, because the
lexical RP is not in the AFD, and in the ‘backwards’ coreference case, (a′),
condition (20b) is met, because the pronoun is not in an argument position;
there is also a clause boundary between them. In (22b, b′) there is again a PP

Influence on choice of template
Syntactic LSC syntactic template

focus structure Cognitive model
inventory PSA [+prag. infl] (in some languages) of context
s Referent-1: Activated
Linking ipant
s o f partic Referent-2: Accessible
algorithm tatu
urse s Referent-3: Inactive
Disco etc.
cal
[do′ (x, …)] CAUSE [BECOME pred′ (y, z)]
n lexi
ce o es
uen c
Infl choi
Lexicon
Figure 12.17 The pervasive role of discourse-pragmatics in grammar
containing a possessive RP, but the PP is in the PrCS, not the PrDP, and this
means the possessor in the RP is in the AFD. Consequently, (b) violates
condition (20a), and coreference is impossible. In (b′), on the other hand,
condition (20a) is met, because the AFD is in the PrCS, and condition (20b)
does not apply because the pronoun is not in an argument position.
These three phenomena all involve IS and the interpretation of sentences,
which is consonant with the idea that the point of IS is to facilitate the
assimilation of information.
12.7 Conclusion
RRG began as a theory of grammatical relations back in the late 1970s, and
one of the fundamental claims made then was that the interaction of
discourse-pragmatics and semantic roles in the constitution of grammatical
relations is variably grammaticalized across languages, and these differences
in grammaticalization underly important typological differences among
languages. While in the ensuing decades the scope of the investigations
carried out in RRG expanded to more and more phenomena, the fundamen-
tal insight that grammatical systems crucially involve discourse-pragmatics
continues to be at the heart of RRG analyses, and this chapter, along with
those by Bentley, LaPolla and Shimojo, as well as the grammatical sketches
in Part V, illustrate this. The idea that discourse-pragmatics permeates gram-
mar is summed up in Figure 12.17, from Van Valin (2005: 182).
References
Balogh, Kata. 2021. Additive particle uses in Hungarian: A Role and Refer-
ence Grammar account. Studies in Language 45: 428–469.

Bentley, Delia. 2008. The interplay of focus structure and syntax: Evidence
from two sister languages. In Van Valin (ed.), 263–284.
from Italo–Romance. Linguistics 56: 1245–1301.
Berio Leda, Anja Latrouite, Robert D. Van Valin, Jr. and Gottfried Vosgerau.
2017. Immediate and general common ground. In Patrick Brézillon, Roy
Turner and Carlo Penco (eds.), Modeling and Using Context, 1–14. Heidelberg:
Springer International Publishing.
Bickerton, Derek. 1975. Some assertions about presuppositions and pro-
nominalization. Communication and Linguistics Studies 11, Parasession on
Functionalism, 580–609.
Bolinger, Dwight. 1979. Pronouns in discourse. In T. Givón (ed.), Syntax and
Semantics, Vol. XII: Discourse and Syntax, 289–310. New York: Academic
Press.
Bresnan, Joan and Sam A. Mchombo. 1987. Topic, pronoun and agreement in
ˇ
Chichewa. Language 63: 741–782.
Demuth, Katherine. 1989. Maturation and the acquisition of the Sesotho
passive. Language 65: 56–80.
Demuth, Katherine and Mark Johnson.1989. Interaction between discourse
functions and agreement in Setswana. Journal of African Languages and
Filchenko, Andrey. 2007. Aspect [sic] of the Grammar of Eastern Khanty. PhD
dissertation, Tomsk State Pedagogical University.
Gundel, Jeanette. 1988. Universals of topic-comment structure. In M. Ham-
mond, E. Moravcsik and J. Wirth (eds.), Studies in Linguistic Typology,
Gundel, Jeanette. 2012. Pragmatics and information structure. In Keith Allan
and Kasia Jaszczolt (eds.), The Cambridge Handbook of Pragmatics, 585–598.
Han, Jeonghan. 1999. On Grammatical Coding of Information Structure in Korean:
A Role and Reference Grammar Account. PhD dissertation, University at Buf-
falo (SUNY) [available on RRG website, https://rrg.caset.buffalo.edu/].
Kluwer.
Kibrik, Alexander E. 1979. Canonical ergativity and Daghestan languages. In
Frans Planck (ed.), Ergativity, 61–78. London: Academic Press.
Kihara, C. Patrick. 2017. Aspects of G~ıkuy
~ u~ (Kikuyu) Complex Sentences: A Role and
Reference Grammar analysis. PhD dissertation, Heinrich Heine University

Kim, Alan H. O. 1988. Preverbal focusing and type XXIII languages. In
M. Hammond, E. Moravcsik and J. Wirth (eds.), Studies in Syntactic Typology,
Koshkaryova, N. 2000. Sentences with ditransitive verbs in Khanty. Paper
presented at the 2000 RRG Conference, Dubrovnik, Croatia.

Krifka, Manfred and Renate Musan (eds.). 2012. The Expression of Information
Structure. (The Expression of Cognitive Categories 5). Berlin: Mouton de
Gruyter.
Kuno, Susumu. 1972a. Functional Sentence Perspective: A case study from
Japanese and English. Linguistic Inquiry 3: 269–320.
Kuno, Susumu. 1972b. Pronominalization, reflexivization, and direct dis-
course. Linguistic Inquiry 3: 161–196.
Kuno, Susumu. 1975. Three perspectives in the functional approach to
syntax. CLS Parasession on Functionalism, 276–336.
Kuno, Susumu. 1991. Remarks on quantifier scope. In H. Nakajima (ed.),
Current English Linguistics in Japan, 261–287. Berlin: Mouton de Gruyter.
Kuno, S., K. Takami and Y. Wu. 1999. Quantifier scope in English, Chinese
and Japanese. Language 75: 63–111.
Lambrecht, Knud. 1987. Sentence focus, information structure, and the
thetic-categorial distinction. Berkeley Linguistic Society 13: 366–382.
Lambrecht, Knud. 1994. Information Structure and Sentence Form. Cambridge:
Latrouite, Anja. 2011. Case and Voice in Tagalog. PhD dissertation, Heinrich
Heine University Düsseldorf [available on RRG website, https://rrg.caset.
buffalo.edu/].
Latrouite, Anja. 2021. Specification predication: Unexpectedness and cleft
constructions in Tagalog. Faits de Langues 52(1): 227–254. doi: https://doi
.org/10.1163/19589514–05201011.
Latrouite, Anja and Arndt Riester. 2018. The role of information structure
for morphosyntactic choices in Tagalog. In Sonja Riesberg and Asako
Shiohara (eds.), Perspectives on Information Structure in Austronesian Lan-
guages. Studies in Diversity Linguistics, 247–284. Berlin: Language Sci-
ence Press.
Latrouite, Anja and Robert D. Van Valin, Jr. 2021. An RRG account of aspects
of the information structure-syntax interface in Tagalog. In Van Valin
(ed.), 257–283.
Levelt, Willem. 1989. Speaking. Cambridge, MA: MIT Press.
Nuhn, Patrick. 2021. Ay-Inversion in Tagalog: Information Structure and Morpho-
syntax of an Austronesian Language. Düsseldorf: Düsseldorf University Press.
O’Connor, Rob. 2008. A prosodic projection for Role and Reference Gram-
mar. In Van Valin (ed.), 227–244.
Park, Ki-seong. 1995. The Semantics and Pragmatics of Case Marking in Korean:
A Role and Reference Grammar Account. PhD dissertation, University at Buf-
falo (SUNY) [available on RRG website, https://rrg.caset.buffalo.edu/].
Planck, Frans. 1995. Double Case: Agreement by Suffixaufnahme. New York:
Schachter, Paul. 1976. The subject in Philippine languages: Actor, topic,
actor-topic, or none of the above. In Charles Li (ed.), Subject and Topic,
491–518. New York: Academic Press.

Schachter, Paul and Fe Otanes. 1972. Tagalog Reference Grammar. Berkeley, CA:
Schieffelin, Bambi. 1985. The acquisition of Kaluli. In Dan I. Slobin (ed.), The
Cross-Linguistic Study of Language Acquisition, 525–593. Hillsdale, NJ: Law-
rence Erlbaum.
Sgall, Petr, Eva Hajicova and Jarmila Panevova. 1986. The Meaning of the
Sentence in Its Semantic and Pragmatic Aspects, ed. by Jacob L. Mey. Dordrecht:
D. Reidel.
numeral quantifier construction in Japanese. In Van Valin (ed.),
285–304.
Van Hooste, Koen. 2018. Instruments and Related Concepts at the Syntax–Semantics
Interface. Düsseldorf: Düsseldorf University Press.
Van Hooste, Koen. 2021. A cross-linguistic survey of the instrument-subject
alternation. In Van Valin (ed.), 169–188.
Van Valin, Robert D., Jr. 1990. Functionalism, anaphora and syntax. (Review
article on S. Kuno, Functional Syntax.) Studies in Language 14: 169–219.
Van Valin, Robert D., Jr. 1994. Extraction restrictions, competing theories
and the argument from the poverty of the stimulus. In Susan Lima,
Roberta L. Corrigan and Gregory K. Iverson (eds.), The Reality of Linguistic
Rules, 243–259. Amsterdam: John Benjamins.
mechanisms of language acquisition. In Michael Tomasello (ed.), The
New Psychology of Language: Cognitive and Functional Approaches to Language
Structure, 221–249. Mahwah, NJ: Lawrence Erlbaum [available on RRG
website, https://rrg.caset.buffalo.edu/].
Van Valin, Robert D., Jr. 2008. Some remarks on Universal Grammar. In J.
Guo, E. Lieven, S. Ervin-Tripp, N. Budwig, S. őzçalişkan and K. Nakamura
(eds.), Crosslinguistic Approaches to the Psychology of Language: Research in the
Tradition of Dan Isaac Slobin, 311–320. New York: Psychology Press.
Van Valin, Robert D., Jr. 2009. Privileged syntactic arguments, pivots and
controllers. In Lilián Guerrero, Sergio Ibáñez Cerda and Valeria A. Belloro
(eds.), Studies in Role and Reference Grammar, 45–68. Mexico: UNAM.
Amazonian languages. In M. M. Jocelyn Fernandez-Vest and Robert D. Van
Valin, Jr. (eds.), Information Structuring of Spoken Language from a Crosslinguis-
tic Perspective, 77–92. Berlin: Mouton de Gruyter.
Van Valin, Robert D., Jr. (ed.). 2021. Challenges at the Syntax–Semantics–
Pragmatics Interface: A Role and Reference Grammar Perspective. Newcastle upon

von Heusinger, Klaus. 1999. Intonation and Information Structure. Habilita-
tionsschrift, University of Konstanz.
Notes
1 The background refers to the non-focal part of the utterance, namely the
two IUs outside of the AFD in Figure 12.2. It may or may not be part of the
pragmatic presupposition, although it usually is.
2 Focus–background and topic–comment represent two logically distinct
types of givenness: relational givenness in the former and referential
givenness in the latter (Gundel 2012).
3 The propositional representations are simplified for the sake of the
discussion; in a more technical presentation they would be represented
by the logical structures introduced in Chapters 1, 3 and 4.
4 For a pro-less analysis of the Mandarin phenomena discussed in Van Valin
and LaPolla (1997) mentioned earlier, see Van Valin (2005: 174–175).
5 Japanese verbs of giving and receiving are very complex with respect to
the social relationships and social statuses of the giver and the receiver.
None of that is relevant to this example, and therefore the verb of
receiving will be represented by an abstract ‘GET’ predicate.
6 See Van Valin (2009) for discussion of two languages (Liangshan Nuosu
[Lolo-Burmese] and Barai [Papuan]) in which IS plays an important role in
PSA selection, despite the lack of formal voice constructions in them.
7 This construction is often referred to as the ‘instrument subject alterna-
tion’. It is not possible in all languages, and in languages which have it, it
is highly constrained. See Van Hooste (2018, 2021) for a detailed discus-
sion of this construction.
8 See Nuhn (2021) for an in-depth RRG analysis of ay-inversion.
9 With intransitive verbs of motion, e.g. kan- ‘go’ as in Chelswu-ka san-ey kan-
ess-ta [Ch-nom mountain-loc go-pst-dec] ‘Chelswu went to(wards) the
mountain’, assigning the goal argument locative case means that
the moving participant may or may not have made it to the goal, as
the translation indicates. It is possible, however, to replace the locative
with the accusative case, Chelswu-ka san-lul kan-ess-ta, in which case the
sentence means that the moving participant did accomplish reaching
the goal, i.e. ‘Chelswu went to(*wards) the mountain’. Obviously, this
replacement of the locative by the accusative cannot be considered case
‘spreading’, since there is no other accusative RP in the clause. Rather, it
is better to speak of ‘case substitution’, which applies equally to this
example and to (17b).

10 Case stacking, also known as Suffixaufnahme, occurs in a variety of lan-

guages (see Planck 1995), and the most common variety involves the
possessor in a possessive construction bearing both genitive case and the
case assigned to the possessed, the head of the construction. These do not
have the semantic or IS effects of the Korean phenomena. A language
which appears to have case stacking which seems to be at least in part
motivated by IS factors is Eastern Khanty (Uralic; Koshkaryova 2000; but
see Filchenko 2007 for an alternative view).

Part Three
Topics in RRG:
Complex Sentences

13
The Structure and
Semantics of Complex
Sentences
Toshio Ohori
13.1 Introduction
Most, if not all, grammatical theories have some framework for handling
complex sentences. The RRG theory of complex sentences, a.k.a. the
theory of clause linkage, was first systematically laid out in Foley and
Van Valin (1984, henceforth FVV), building on Olson’s work (1981). Since
then, significant revisions have been made, which one may find in Van
Valin and LaPolla (1997, henceforth VVLP) and in Van Valin (2005). The
RRG theory of complex sentences is unique in that, on the one hand, it is
based on the layered representation of clause structure, and, on the other
hand, it goes beyond the traditional coordination–subordination dichot-
omy, so that the theory provides a comprehensive framework which
captures both typological diversity and various facets of diachronic devel-
opment. In this chapter, we review the RRG theory of complex sentences
drawing upon a wide variety of languages. Due attention will be paid to
some of the recent developments both inside and outside the RRG
framework.
In the remainder of this section, the RRG theory of sentence structure
will briefly be reviewed as a backdrop for the later discussion. Section 13.2
will discuss the structural aspects of clause linkage, and Section 13.3 its
semantic aspects. Section 13.4 will provide a sample analysis of complex
sentences within RRG, focusing on the interaction of linkage types, oper-
ators and reference tracking. In Section 13.5 we offer some concluding
remarks.
13.1.1 The RRG Theory of Sentence Structure

RRG belongs to functional linguistics (cf. chapter 1 of VVLP) insofar as it
pays equal attention to both structure and meaning of the sentence.
Within this paradigm, it is assumed that syntactic structures are

526 TOSHIO OHORI
motivated in the sense that they have semantic/pragmatic correlates, and

the way such correlations hold can be formulated within a relatively
narrow range of variation.
In RRG, one basic unit of grammatical analysis is the clause, which is
characterized by a three-layered endocentric structure. The innermost
layer, the nucleus, is typically realized by the verbal predicate which
encapsulates the skeletal structure of an event. The nucleus is then com-
plemented by obligatory arguments, resulting in the core. Finally, the core
plus optional adjuncts (called periphery in RRG) make up the clause. The
clause in itself may or may not constitute a unit of pragmatic force such as
assertion and question. When it does, it is identified as a sentence. Other-
wise, a clause remains just a clause whose pragmatic force is determined
in relation to some other clause which does have such a force. Example (1)
is an illustration of the structure of a simple clause from German. As for
the operator projection (simplified here), which is represented in
Figure 13.1, see (10).
(1) German (Indo-European)

Man sollte natürlich alle Ghibli Filme gesehen haben.
person should naturally all Ghibli films seen have
‘One should have naturally seen all Ghibli films.’
The layered model of the clause is drastically different from the model of
clause structure in many other grammatical theories, which posit the
phrasal category VP. In RRG, there is no room for this category. This way
Figure 13.1 The layered structure of the clause

The Structure and Semantics of Complex Sentences 527
of looking at the clause structure is shared by Functional Discourse Gram-

mar (for example, Hengeveld and Mackenzie 2008), and is preceded, though
remotely, by field-oriented approaches such as tagmemics.
In addition, RRG postulates special structural positions outside the core.
One is the pre-core slot (PrCS), typically reserved for a constituent bearing a
special, marked, informational status, such as the target of content ques-
tions (for example, wh-questions in English) and the pre-core focus. Another
is the dislocated position on either side of the clause periphery, which used
to be called left-detached position (LDP) or right-detached position (RDP),
and is now called pre-detached position (PrDP) and post-detached position
(PoDP). This gives a piece of presupposed information, often characterized as
a sentence topic.
13.2 Structural Types of Clause Linkage
13.2.1 Juncture: The Layer of Linkage

The first parameter for the characterization of complex sentences is the
layer of the clause on which the linkage takes place. This is called juncture.
The second parameter, called nexus, is concerned with the nature of the
linkage, namely the type of dependency that holds between the linked units.
This will be discussed in 13.2.2.
A maximally broad concept of complex sentences would include linking
at any of the three layers of clause structure. We will not exclude, therefore,
complex predicates such as Japanese nomi-hosu ‘drink-dry’ > ‘drink up;
empty’. Such a broad construal of the term is justified on the grounds that
by providing a comprehensive framework for classifying a wide variety of
constructions, we may be able to characterize in a coherent manner the
process of category change from canonical cases of complex sentences to less
canonical ones. Hence, it would be more appropriate to adopt the term
complex structures instead of complex sentences. In what follows, we use
complex structures as a generic term, and limit complex sentences to refer
to clause linkage proper. In tree diagrams, linkage markers are shown as LM
for shorthand.
First, clause juncture, defined as the linkage at the clausal layer, is illus-
trated by the following examples. In this section, we shall limit ourselves to
coordination for simplicity’s sake. Other types of nexus relations on all three
layers will be illustrated in 13.2.2.
(2) Maori (Austronesian; Bauer 1993: 125. ‘>’ indicates the possessor–possessee
relation )
Me haere atu (raanei) ia ki toou kaainga, me haere
aux move away or 3sg to 2sg.gen>sg home aux move
mai raanei koe ki konei
hither or 2sg to here
‘He could come to your house or you could come here.’

528 TOSHIO OHORI
(3) Iraqw (Afro-Asiatic; Mous 1993: 298)

i ‘ayto’ó-r weer<am>iis nee aníng a tlax-úum
sbj.3 maize-f sell<dur>.3sg.m.prs and 1sg obj.f buy-dur.1sg
‘He is selling maize and I am buying it.’
(4) Sonora Yaqui (Ute-Aztecan; Dedrick and Casad 1999: 363)

′aháa n-áčai ′ó′olaa tú’i-si n¼ám yi′í-pea táa nee
aha my-father ancient good-adv I¼them dance-desiderative but I
káa-tú′ii-m wók-ek
neg-good-pl foot-have
‘Aha, my revered Father, how I wish I could dance to that music, but I just
don’t have good legs.’
Languages may vary with respect to the morphological strategy they adopt
for marking coordination. While all coordination markers in (2)–(4) (raanei
in (2), nee in (3) and táa in (4)) are words, a clitic or a coordinating particle
may also be used. Zero-marking is also not infrequent. In the Maori sentence
(2), raanei can be omitted, and in that case the resulting sentence would
allow either a conjunctive or a disjunctive interpretation, depending on
contextual factors. Generally speaking, the coordinating linkage is of a
‘balanced’ type (Croft 2001) and neither of the linked units is formally
reduced or de-ranked. Admittedly, coordinate constructions may have an
asymmetric interpretation such as temporal sequence and cause–result, but
this is usually taken to be an outcome of inference based on frame-semantic
knowledge and contextual information. Schematically, clause coordination
is represented in Figure 13.2. Here, there is no obligatory sharing of an
argument, and the linked clauses may describe separate events, each with
its own Actor-Undergoer assignment.
Next, core juncture is the linking of cores, that is, the predicate plus
obligatory arguments, and it involves partial sharing of argument
Figure 13.2 Clausal coordination in Maori

structure. The shared argument acts as something like a ‘hub’ for joining
two units.
(5) French (Indo-European)

Je laisserai Jean manger les gâteaux.
I will.let John eat the cakes
‘I’ll let Jean eat the cakes.’
(6) Thai (Tai-Kadai; Diller 2006: 172)

bo’risa:t1 nam sin4kha:3 khaw3
company lead product enter
‘The company imports products.’
(7) Turkish (Turkic; Watters 1993: 550)

Odam-da yat-mış uyu-ya-mı-yor-du-m.
room-loc lie-aux sleep-aux-neg-prog-pst-1sg
‘I was lying in my room unable to go to sleep.’
Core juncture consists of mainly two subtypes depending on the mode of

argument linking. Examples (5)–(6) represent the first type, where the shared
argument in each sentence (Jean in (5) and in sin4kha:3 in (6)) serves two
different functions, namely Undergoer and Actor, in the linked cores.
Example (7) represents another type, where the shared argument plays the
role of Actor in both of the linked cores. In those theories that posit VP, (7)
would be analysed as VP conjoining.
The schematic representation of the structure of (5) is shown in
Figure 13.3. While the macrorole of the shared core argument is different
between (5)–(6) and (7), the fundamental structure is the same, and syntax–
semantics mapping is handled by the independent linking algorithm in
combination with the logical structure of the predicate. Besides the sharing
of an argument, core juncture (in the present case core coordination)
Figure 13.3 Core coordination in French

530 TOSHIO OHORI
involves obligatory sharing of a periphery, which is a natural consequence of

the notion of periphery as an element external to the core.
Third, nuclear juncture is formed on the nucleus layer. Here too, a rela-
tively simple example of nucleus coordination is given in (8).
(8) Tukang Besi (Austronesian; Donohue 1999: 182)

No-helo’a-ako te ana-no te roukau
3realis-cook-do.for core child-3poss core vegetables
‘She cooked the vegetables for her children.’
In this example, the linked nuclei, helo’a ako ‘cook-do.for’ act like a simple,
unitary predicate, and all arguments are shared. The nucleus-level
coordination may be represented as in Figure 13.4.
The formation of a unitary predicate is a prominent characteristic of the
nucleus juncture. Compare (8) with (9).
(9) Tukang Besi (Donohue 1999: 182)

No-helo’a te roukau ako te ana-no
3realis-cook core vegetables do.for core child-3poss
‘She cooked the vegetables for her children’
While the same predicates occur in (8) and (9), the linkage structures are
different. In the former, two core arguments (marked by the non-nominative
core article te) follow the unitary predicate which acts like a ditransitive verb.
In (9), which is an instance of core juncture, each predicate takes its own
Undergoer core argument while sharing the Actor. In the linguistic literature,
the term serial verb construction (SVC) is often used inconsistently, but the
RRG typology of complex structures enables us to distinguish explicitly
between the linkage at the core level and that at the nucleus level.
In addition to the layered structure of the clause, RRG has other levels of
representation, namely the operator projection, information structure,
Figure 13.4 Nuclear coordination in Tukang Besi

and logical structure. These systems interact with clause linkage in various
ways. Here, we shall look at the projection of operators in complex struc-
tures. According to Van Valin (2005: 9), the following operators can be
identified.
(10) Nucleus:
Aspect
Negation
Directionals (only those modifying orientation of action or event without
reference to participants)
Core:
Directionals (only those expressing the orientation or motion of one
participant with reference to another participant or to the speaker)
Modality (root modals)
Internal negation
Clause:
Tense
Evidentials
Illocutionary force
When they occur in a coordinate linkage, operators that belong to the layer
higher than the layer where the linkage occurs necessarily take both linked
units within their scope. For example, clausal operators such as tense are
always shared by the linked cores (and, naturally, by the linked nuclei, as
can be seen in (8) and (9)). Core operators, for example root modality, are
always shared by the linked nuclei. An illustration of this interaction in the
Turkish example (7) is given in Figure 13.5. In linkages other than coordin-
ation, different situations hold, which we will discuss in the next section.
13.2.2 Nexus: The Type of Dependency

Traditionally, only two types of structural relationship have been identified
in order to characterize the linking of predicative units, namely coordin-
ation and subordination. Coordination involves no dependency; two units
are juxtaposed with or without an overt marker for the coordinate relation.
Subordination involves embedding: a structural unit (clause, core or
nucleus, in our terms) serves as a constituent dominated by some higher
unit (we shall elaborate on this definition below).
Let us start with the linking of clauses. We have already seen instances of
clausal coordination (2)–(4), where there is no structural dependency. Also, all
operators save the outermost one (i.e. illocutionary force) are specified inde-
pendently for each linked unit, whether they are explicitly marked or not.
In contrast, subordination involves embedding, but here one termino-
logical clarification is in order. When we say that a certain sentence
exemplifies clausal subordination, the term ‘clause’ is applied to the

532 TOSHIO OHORI
Figure 13.5 Core coordination in Turkish with operator projection
subordinated unit. For example, a clause may occupy an argument position

within a core, as in (11), but we do not call it core subordination. When we
refer to an instance of X-to-Y linkage (in this case clause-to-core), we call it by
the layer on which X is formed. Hence (11) is an instance of clausal subordin-
ation, in which the clause achka chaya-mu-sha-n-ta ‘so many arrived’ occupies
a core argument position of the predicate musya-shka-:-chu ‘I did not know’,
functioning as a complement clause.
(11) Huallaga Quechua (Quechuan; Weber 1989: 289)

Mana musya-shka-:-chu achka chaya-mu-sha-n-ta.
not know-prf-1-neg many arrive-afar-nmlz-3-obj
‘I did not know that so many arrived.’
There are two other types of clausal subordination. A clause may occur in
an adverbial periphery modifying a clause, as in (12) (i.e. clause-to-periphery
or ad-clausal subordination), or it may serve as a constituent modifying a
noun, as in (13) (i.e. clause-to-noun subordination) forming a relative clause.
(12) Huallaga Quechua (Weber 1983: 112)

noqa eskuyla-chaw ka-yka-sha-:-pita aywa-ku-ra-: Tingo Maria-pa
I school-loc be-ipfv-nmlz-1-abl go-refl-pst-1 Tingo Maria-gen
‘From (the circumstance of ) being in school (i.e. after it, from that place, and
being tired of being in school) I went to the Tingo Maria area.’

(13) Huallaga Quechua (Weber 1989: 280)

Maqa-shu-q runa sha-yka:-mu-n.
hit-2obj-nmlz man come-ipfv-afar-3
‘The man who hit you is coming.’
In Huallaga Quechua, subordination is typically marked by a nominalizer

(called a ‘substantivizing’ marker by Weber and glossed as nmlz in the
above example), which is -sha in (11)–(12) and -q in (13). In (12), the clause
noqa eskuyla-chaw ka-yka-sha-:-pita ‘From (the circumstance of ) being in school’
bears the nominalizer sha, which in turn is followed by the ablative case
suffix pita, forming an adverbial periphery. In (13), the clause maqa-shu-q
‘(someone) who hit you’ bears the agentive nominalizer, and it modifies the
noun runa ‘man’, forming a larger RP. Figures 13.6–13.8 are structural
representations of the said three types of clausal subordination.
Subordination can also be formed on core and nuclear layers. Core-to-core
subordination is illustrated by (14).
(14) David regretted Amy’s losing the race. (VVLP: 455)
Here, the constituent Amy’s losing the race is embedded in the argument position
of another core, David regretted. The subordinated core is in participial form and
thus devoid of tense marking, which is a clause-level operator.
Next, example (15) illustrates nucleus-to-nucleus subordination, where the
second nucleus tέ ‘stay’ adds an aspectual meaning (namely progressive) to
the first nucleus gàrá ‘write’.
(15) Khwe (Khoisan; Pavey 2010: 234)

Xàmá thám̀ à kgàrá-ná tέ-ὲ-tè.
3sg.m letter obj write-link stay-link-prs
‘He is writing a letter.’
Figure 13.6 Clausal subordination (clause-to-core/complement clause) in Huallaga Quechua

534 TOSHIO OHORI
Figure 13.7 Clausal subordination (clause-to-periphery/adverbial clause) in

Huallaga Quechua
Figure 13.8 Clausal subordination (clause-to-noun/relative clause) in

Huallaga Quechua
Since the nucleus does not have an argument position by definition, the
subordinate relation is that of adjunction, not embedding. In (15), the
second nucleus tέ ‘stay’ can be seen as an operator by itself, given its
aspectual meaning. This type of nuclear juncture is dubbed ‘operator con-
struction’ by Hasegawa (1996) for analogous cases in Japanese.
The constituent structures of core subordination (core-to-core embedding)
and nuclear subordination (nucleus-to-nucleus adjoining) are shown in Fig-
ures 13.9 and 13.10 respectively.

Figure 13.9 Core subordination in English
Figure 13.10 Nuclear subordination in Khwe (Pavey 2010: 235)
Besides coordination and subordination, we may identify yet another type

of structure which does not fall under either category. Consider (16), which
is also a linking on the clausal layer.
(16) Udihe (Tungustic; Nikolaeva and Tolskaya 2001: 790)

Eke bi-lisi-ni ’ana-wa aisi-zeŋe-i
time be-cvb-3sg boat-acc mend-fut-1sg
‘If I have time (if there is time), I will mend the boat.’
Here, the nexus type is distinguished from both coordination and subordin-
ation by the following properties. First, the first clause, bi-lisi-ni ‘(there) is
time’ is devoid of clause-level operator marking, while the second clause
aisi-zeŋe-i ‘I will mend the boat’ is marked for tense (namely future). The
form which indicates linkage, namely lisi, is a subtype of non-finite ending,

536 TOSHIO OHORI
that is, it belongs to a conditional converb class. Second, the first clause is
not governed by any overt subordinating marker, nor does the linkage
involve embedding as in the case of subordination. Linkage like that in
(16) often results in a relatively long clause chain, a characteristic shared
by coordination but not by subordination. Third, the non-finite verb form
indicates the change of subject (notice that the first clause has the syntactic
subject eke ‘time’ and the second clause first-person singular), which is not
an available morphological option for the finite clause in this language.
In order to capture this type of nexus relation, that is, linkage involving
operator dependency but not embedding, RRG adopts the term
cosubordination, after the work by Olson (1981) on Barai. In addition to
these general properties, Roberts (1988) argues, based on evidence from
Amele, that cosubordination is differentiated from subordination on the
following grounds: (i) a subordinate clause can occur either initially or
finally with respect to the main clause, which is not possible for a
cosubordinate clause; (ii) the way pronominal reference works is different
in cosubordinate and subordinate clauses; and (iii) co-occurrence restric-
tions on conjunctive particles are different between subordinate and
cosubordinate clauses.
Admittedly, cosubordination is a highly schematic category and languages
may exhibit different concomitant properties within this broad definition.
The schematic structure of clausal cosubordination (16) is represented in
Figure 13.11. The operator projection is also given in order to visualize
clause-level operator dependency. It is true that not all cosubordinate con-
structions allow clause chaining and switch-reference marking, though they
are common among Papuan languages. However, the introduction of the
third type of nexus relation enables us to analyse clause linkage phenomena
in a more comprehensive and precise manner.
Cosubordination on the core layer is exemplified by (17). This is known as
a control construction, where the ‘missing subject’ of try is linked to the
‘higher subject’ John.
(17) John must try to wash the car.
This sentence entails that John must wash the car. Hence the second core
wash the car is operator-dependent on the first core for deontic modality
expressed by must (VVLP: 460). Compare (17) with core coordination, exempli-
fied by (18).
(18) John must tell Bill to wash the car.
From this sentence, it does not follow that Bill must wash the car, that is,
the second core wash the car is not operator-dependent on the first core. In
other words, the root modality operator must has scope only over the
first core.
Finally, in nuclear cosubordination, two linked nuclei form a unitary
predicate like nuclear coordination. Compare (19) with its core-level

Figure 13.11 Clausal cosubordination in Udihe with operator projection
counterpart (5). Unlike in core juncture, where the causee Jean occurs as the
‘hub’ for both linked cores, here this argument is expressed as an oblique
constituent marked by the preposition à (see also the Tukang Besi examples
(8) and (9)).
(19) French
Je ferai manger les gateaux à Jean.
I will.make eat the cakes to Jean
‘I will make Jean eat the cakes.’
It is not always easy to distinguish between coordination and cosubordi-

nation at the nuclear level, because there are not many types of operators
that apply to this layer. One candidate, however, is aspect. The Japanese
example (20) illustrates the point.
(20) Japanese (isolate)

Keezai-ga sakamiti-o korogari-oti-tei-ru.
economy-nom downhill-acc roll-fall-prog-prs
‘The economy is plummeting (lit. roll-falling the downhill).’
In this example, two nuclei, namely korogari- ‘roll’ and oti- ‘fall’, are linked.
The aspect marker -tei- occurs next to the second unit, but it modifies both
nuclei. (20) entails that the economy is (metaphorically) rolling-and-falling

538 TOSHIO OHORI
Figure 13.12 Core cosubordination in English
Figure 13.13 Nuclear cosubordination in French
downhill, and in this sense the first nucleus is operator-dependent on the

second nucleus.
The structural representations for core cosubordination and nuclear
cosubordination are given in Figures 13.12 and 13.13.
The three nexus types which we have identified so far are characterized by
the distinctive feature matrix specifying the kind of dependency involved in
the linkage.
(21) coordination [distributional, operator]

subordination [þdistributional, operator]
cosubordination [þdistributional, þoperator]
Coordination does not involve either kind of dependency. When two pre-
dicative constituents are coordinated, either can stand by itself. Subordin-
ation involves distributional dependency only; a subordinate constituent

can take its own operator that applies to the layer of linkage. Cosubordina-
tion involves both types of dependency, and, in this sense, it can be seen as
the most tightly integrated nexus relation. It is generally known that subor-
dination can be divided into two types, namely structures that involve
embedding in an argument position and those that do not (i.e. adjunction).
This matrix can therefore be further elaborated by introducing the feature
[argument] to capture these two subtypes of subordination. Since it is a
reasonable assumption that the argument position is more tightly inte-
grated into the clause structure than the adjunct position, we may arrange
the two subtypes of subordination in the following way:
(22) coordination [distributional, operator]

subordination [þdistributional, operator, argument]
[þdistributional, operator, þargument]
cosubordination [þdistributional, þoperator]
A general summary of linkage types will be given in Section 13.3.
13.2.3 Elaborations on the Basic Matrix

Since the typology of clause linkage was initially introduced in RRG, some
important revisions have been made. We shall review two of them, mainly
drawing upon Van Valin (2005) and Pavey (2010).
The first revision concerns juncture on the sentence level, that is, the
combining of sentences having independent illocutionary force. RRG draws
a clear distinction between clauses and sentences, so this issue is more than
a terminological matter. In this domain, only coordination and subordin-
ation exist. Sentence cosubordination is non-existent, because the defining
feature of this type of nexus is operator dependency, and it is simply
contradictory for a sentence, defined as a unit with full specification of
operators, to have operator dependence.
Sentence coordination is exemplified by (23) from English, where two
sentences, one an imperative and the other a declarative, are combined. In
English orthography, this is a single ‘sentence’, but from a theoretical point
of view, it is a mini-discourse consisting of two sentences (note that (23) can
be rewritten as Be quiet! Or I’ll call the police without relevant change in
meaning). The structure of (23) is illustrated in Figure 13.14. One could say
that be quiet is not really an imperative but a conditional protasis, meaning
‘if you do not become quiet’, but what matters in this analysis is the scope of
illocutionary force, whereas the conditional meaning is derived from the
entire construction, which should be specified separately.
(23) Be quiet, or I’ll call the police.
As for sentence subordination, there are multiple possibilities. Parenthet-

ical clauses are one candidate (another type of sentence subordination will
be introduced below when we discuss the information structure of the

540 TOSHIO OHORI
Figure 13.14 Sentential coordination in English
Figure 13.15 Sentential subordination (parenthetical) in English (Kaltenböck et al. 2011: 849, orig.
from Burton-Roberts 2005: 180)
sentence). (24) is an instance of sentence-to-sentence subordination and a

structural representation is also provided in Figure 13.15.
(24) The main point – why not have a seat? – is outlined in the middle paragraph.
The second recent revision in the typology of clause linkage concerns the
elaboration of the layered structure of the clause (see Bickel 1993; Van
Valin 2005). In addition to the initial formulation by FVV, RRG has accom-
modated the left-detached position (LDP), now called the pre-detached
position (PrDP), and the pre-core slot (PrCS) in order to capture the infor-
mation structure of the sentence. Let us see how they interact with clause
linkage in turn.
Generally speaking, the PrDP is reserved for providing a background
assumption, and is separated from the main clause by a prosodic break.

A sentence is allowed to have only one PrDP, so its presence counts as a

criterion for sentencehood. The PrDP position is often occupied by a PP
expressing a spatio-temporal setting, but it can also host a predicative unit,
as in the French example (25).
(25) French (Van Valin 2005: 192)

Moi, quand j’étais jeune, on parlait seulement en français.
I when I was young one spoke only in French
‘Me, when I was young, we only spoke in French.’
Here, the subordinate unit, moi, quand j’étais jeune ‘me, when I was young’,
has ‘its own left-detached expression, which makes it a sentence, and this
sentence is then in the left-detached position of the matrix sentence’ (Van
Valin 2005: 192). In this sense, (25) illustrates another type of sentence
subordination, namely sentence-to-PrDP subordination.
Next, the PrCS is distinguished from the PrDP in that it is a subpart of the
clause which is reserved for a marked or narrow focus and is not followed by
a prosodic break. The PrCS is typically occupied by an RP, but a clause may
also occur in this position (the sentence cannot occur in the PrCS by defin-
ition). Unlike in the case of sentential juncture, a clause which occurs in the
PrCS is devoid of illocutionary force. The contrast between sentence-to-PrDP
subordination and clause-to-PrCS subordination may be illustrated by the
following pair of sentences from Japanese.
(26) Japanese
Sainoo-ga ar-u-nda-kara ganbar-e.
talent-nom have-prs-cop-because work.hard-imp
‘Because/since (you) have a talent, work hard.’
(27) Japanese
Sainoo-ga ar-u-kara koso seekoo-si-tei-ru.
talent-nom have-prs-because foc success-do-res-prs
‘Precisely because/??since (X) has a talent (s/he) has made a success.’
In (26), the first predicative unit ends with the assertive predication marker
nda. The kara-marked unit expressing reason gives a presupposed piece of
information, and though the main clause is in the imperative form, this
presupposition remains unchallenged. The fact that the scope of the impera-
tive is limited to the second predicative unit supports the analysis of (26) as
an instance of sentence-to-PrDP subordination. In contrast, in (27) the link-
age marker kara is followed by the focus marker koso, indicating that sainoo-
ga ar-u ‘(X) has a talent’ occupies the PrCS, and hence (27) is an instance of
clause-to-PrCS subordination. Without the focus marker koso, a default
analysis of this sentence would be clause-to-periphery (¼ad-clausal) subor-
dination. As expected, inserting the assertive predication marker -nda before
kara in (27) results in a highly unnatural sentence because of the clash in
information structure, -nda strongly preferring to occur in the PrDP and koso
in the PrCS.

542 TOSHIO OHORI
Figure 13.16 Sentential subordination (PrDP) in Japanese
Figure 13.17 Clausal subordination (PrCS) in Japanese
(28) ??Sainoo-ga ar-u-nda-kara koso seekoo-si-tei-ru.

talent-nom have-prs-cop-because foc success-do-res-prs
The structures of (26) and (27) would look like Figures 13.16 and 13.17.
So far, we have identified six types of subordination involving either a
clause or a sentence, namely clause-to-core (11), clause-to-periphery (¼ad-
clausal) (12), clause-to-noun (13), sentence-to-sentence (24), sentence-to-PrDP
(25)–(26), and clause-to-PrCS (27). In addition, Van Valin (2005) introduces
another kind of clause-to-periphery subordination, following the revision of
RRG to admit a periphery modifying the core, as in the following example
(Van Valin 2005: 194).

Figure 13.18 Clausal subordination (clause-to-core periphery/ad-core) in English
(29) Kim saw Pat after she arrived at the party.
Here the clause after she arrived at the party is analysed as modifying the
core (as shown in Figure 3.18). A semantic motivation for analysing this
example in this way is that the specification of time is more relevant to the
propositional content of the clause compared to the semantic relations
expressed by markers such as although, because, and if, all of which have
more epistemic orientations. To keep this type of clause-to-periphery subor-
dination distinct from the structure like (12), the term ad-core
subordination is introduced. Examining to what extent this subdivision of
clause-to-periphery subordination is viable in the light of typological data
will be a topic for future investigation.
Besides the introduction of sentence-level junctures and the addition of
PrDP and PrCS as hosting positions for linkage, there have been debates on
the status of cosubordination. Most notably, Bickel (2010) and Foley (2010)
independently advance arguments against the notion of cosubordination.
Due to the limitation of space, we will not examine the validity of their
arguments here, but see Van Valin (2021) for a counter-argument.
13.3 The Semantics of Clause Linkage and Form–Meaning

Correlations
The linkage types we have identified thus far can be listed as follows, in
order of stronger to weaker structural integration. For the sake of simplicity,
cosubordination is placed above subordination for each juncture type, but
the order between them is rather tentative compared to the relative order of
other relations.

544 TOSHIO OHORI
(30) Stronger integration

nuclear cosubordination
nuclear subordination
nuclear coordination
core cosubordination
core subordination
core coordination
clause cosubordination
clause subordination
clause-to-core
clause-to-noun
clause-to-periphery (ad-core)
clause-to-periphery (ad-clausal)
clause-to-PrCS
clause coordination
sentence subordination
sentence-to-PrDP
sentence-to-sentence
sentence coordination
Weaker integration
Before we discuss the semantics of complex constructions, we shall intro-

duce one conceptual clarification. The linkage types given in (30) are not
grammatical constructions in Construction Grammar terms (Fillmore et al.
1988; Goldberg 1995). Rather, they are language-independent structural
schemas abstracted from existing sentence types in a wide variety of lan-
guages. In RRG terms, constructions are instead defined with reference to
specific instantiations of structural schemas in a particular language and
the semantic information associated with them.
One of the important insights of RRG, and functional linguistics more
broadly, is that there is a systematic correspondence between form and
meaning (sometimes discussed under the term iconicity; cf. Givón 1980,
1985; Haiman 1985). In this particular case, the generalization is that the
relative strength of unit integration on the structural side correlates with
the strength of integration on the semantic side. Then, the question is how
we might represent the hierarchical arrangement of semantic relations.
Silverstein (1976, 1993) proposed the following list as ‘logical relation of
clauses’ (Silverstein 1993: 481).
(31) possessive
habitual actor
habitual agent
relative clause (making definite reference)
purposive complement
desire complement
indirect discourse complement
temporal adverbial clause
if–then

disjunction
conjunction
clause sequence (sequitur)
clause sequence (non-sequitur)
Silverstein’s idea was incorporated into FVV, and since then the list of
semantic relations, called interclausal semantic relations, has been updated.
Ohori (2001, 2005) proposed that the list of semantic relations should be
made more systematic in a way that derives certain relations from other
more primitive concepts. He introduced a feature system based on a small
number of conceptual distinctions, for example, [þ/ control], [þ/ action
coherence], and [þ/ temporal proximity], which capture stages of concep-
tual elaboration. This idea was partly adopted by Van Valin (2005: 206–207),
where the following list of semantic relations is proposed.
(32) Stronger integration

Causative [1]
Phase
Modifying subevents:
Manner
Motion
Position
Means
Psych-action
Purposive
Jussive
Causative [2]
Direct perception
Indirect perception
Propositional attitude
Cognition
Indirect discourse
Direct discourse
Circumstances
Reason
Conditional
Concessive
Temporal
Simultaneous states of affairs
Sequential states of affairs
Temporally unordered states of affairs
Weaker integration
When the two hierarchies (30) and (32) are aligned, we obtain what is called
the Interclausal Relations Hierarchy (IRH). Naturally, languages differ with
respect to the types of commonly employed linkage types, and the range of
semantic relations that constructions of a given linkage type serve to express
also differs from language to language. Hence the typological generalization
RRG makes about the form–meaning correspondence in the realm of

546 TOSHIO OHORI
Figure 13.19 Form–meaning correlation in complex structures
complex construction is: when a construction that belongs to a certain

syntactic linkage type is associated with some semantic relation, construc-
tions that belong to linkage types on the higher positions on the hierarchy
are associated with semantic relations that are also higher on the hierarchy.
That is, the correspondence between the two hierarchies will look like
Figure 13.19a, but not like 13.19b.
Let us substantiate the claim of the IRH by first looking at the expression
of causal relations. Given two states of affairs X and Y, the relation X CAUSE
Y may hold in a variety of ways along several conceptual parameters. For
example, Causative [1] entails X’s direct manipulation and the accomplish-
ment of the resulting state Y, while Causative [2] and Jussive do not have
these properties. The former type of causative relation tends to be encoded
either morphologically or by nuclear juncture (if not by lexical verbs), but
the latter types of relation tend to be encoded by either core or clause
juncture, that is, by periphrastic causatives (cf. Shibatani’s 1972 classical
discussion of kill and cause to die).
This point is clearly illustrated by the following examples from Cora. (33)
and (34) are examples of morphological and periphrastic causatives
respectively.
(33) Cora (Uto-Aztecan; Vásquez Soto 2002: 225)

í̵ Joel ácipoʔu pú wa-ta-ráʔaraʔi-te.
det Joel butterfly sbj.3sg compl-prf-fly-caus
‘As for Joel, he made the butterfly fly.’
(34) Cora (Vásquez Soto 2002: 232)

í̵ Juan ru-yáuh pu wa-taʔáih ti wáka-si
det John poss.refl.3sg-son sbj.3sg compl-send sbj.comp.3sg cow-pl
wa-náwaʔa-n
compl-steal-irr
‘As for John, he sent his son to steal the cattle.’

In example (33), the causative is formed by the addition of the causative

suffix -te to the verb stem ráʔaraʔi ‘fly’. Vásquez Soto reports that in Cora
morphological causatives mostly, if not solely, involve non-agentive causees.
Therefore (33) is acceptable only if the causee is taken as an inanimate entity
and causation involves direct manipulation. In contrast, in (34) which is an
instance of periphrastic causative (more specifically clause-to-core subordin-
ation), taʔáih ‘send’ functions as the causative verb and the caused event
appears as a subordinate clause marked by the pronominal clitic ti. Here the
causee has volitional control over his act, unlike in (33) (in fact, the causa-
tivization of agentive intransitive is highly restricted for both types of
causatives in Cora, but we do not discuss this issue here). An accompanying
semantic difference between (33) and (34) is temporal proximity between
cause and result. In (33), Joel’s acting on the butterfly immediately causes its
flying, while in (34) Juan’s acting on his son makes the latter’s stealing
happen after some temporal interval.
Another set of examples illustrating the IRH comes from the expression of
temporal relations. In Japanese, the morpheme nagara expresses simultan-
eity. One of its common uses is the expression of manner, namely modifica-
tion of subevent, as in (35).
(35) Japanese
Kare-wa kutibiru-o yugame-nagara warat-ta
He-top lip-acc twist-while smile-pst
‘He smiled while twisting his lips.’
Structurally, this is an instance of core cosubordination. Hence actor

coreference is obligatory in this construction. However, nagara can also be
used in clause-level linkage, namely clausal cosubordination, as in (36), with
no coreference constraint.
(36) Japanese
Keehoo-ga nat-tei-nagara mina okunai-ni i-ta.
alarm-nom sound-prog-while everyone indoor-dat stay-pst
‘While the alarm kept ringing, everyone stayed indoors.’
In this sentence, nagara also connects two states of affairs that occur simul-
taneously. But unlike (35), the nagara-marked clause does not constitute a
subevent of the main clause, mina okunai-ni ita ‘everyone stayed indoors’.
What is more, (36) allows a concessive meaning besides simultaneity, as one
can imagine from the English gloss. Thus the relative strength of unit
integration on the structural side (namely core vs. clause cosubordination)
correlates with that on the semantic side (namely manner subevent vs.
concessive relation).
While the discussion in this section has been brief, evidence that will
reinforce the IRH can be garnered from a greater variety of languages.
More importantly, we may add a further dimension to the IRH by taking
into consideration more specific morphosyntactic features. In fact, one of

548 TOSHIO OHORI
Silverstein (1993)’s original insights was that the logical relation of

clauses (31) exhibits correlation with a wide range of grammatical phe-
nomena, namely: (i) probability of normal forms, (ii) probability of nom-
inalization, (iii) degree of formal distinctness from unlinked clause,
(iv) markedness of connection, (iv) probability of antipassivization, and
(v) suspension of agent hierarchy. Similar threads of discussion have been
advanced by, among others, Givón (1980), Haiman and Thompson (1984),
and Lehmann (1988). (37) is a summary of morphosyntactic features
that correlate with the tightness of interclausal relations, adapted from
Ohori (2005).
(37) Stronger integration Weaker integration

( )
Nominals
Same subject Different subject
Not realized (e.g. ‘equi-deleted’) Realized
Constrained case marking Normal case marking
Verbals
Reduced inflection Elaborate inflection
Grammaticalized With full lexical content
Voice alternation suspended Voice alternation at liberty
Operators
Shared Not shared
Dependent Not dependent
Others
No explicit signal Explicit signal
Word order fixed Free word order
We may now conceptualize the relative strength of unit integration in

complex structures in a triadic form: a set of schematic linkage types as in
(30), a set of semantic relations as in (32), and a set of morphosyntactic
features as in (37), the last of which being more directly applicable to
specific constructions from individual languages.
13.4 How It All Works: Old Japanese ‘Switch-Reference’

and Beyond
In this section, we shall see how the RRG theory of clause linkage serves
to enhance our understanding of specific grammatical phenomena,
focusing on Old Japanese (hence OJ) ‘switch-reference’ (Ohori 1992,
1994, 2002; see also McAuley 2002). It will be shown that what appears
to be a switch-reference system in OJ is in fact epiphenomenal, and the
apparent switch-reference function derives from the typology of clause
linkage. Put differently, reference tracking is a concomitant feature of
the clause-linkage structure, along with other important grammatical
features.

Let us start with Akiba (1977, 1978)’s study, where the uses of conjunctive
particles in the texts from tenth-century Japanese are closely examined.
Rejecting the idea that these particles should be characterized on semantic
grounds (for example, in terms of condition, cause, adversity, etc.), Akiba
asserts, ‘they are better characterizable in terms of the switch-reference
function’ (1977: 611). In fact, according to Akiba’s calculation based on
Taketori Monogatari, in 94 per cent of 536 occurrences of the linkage by te
the subject is retained across clauses, while in 92 per cent of 120 occurrences
of the linkage by ba there was switching of the subject. The following are
representative examples of linkage by te and ba. Because these markers are
semantically underspecified, they are glossed simply as link. Schematic
representations of the sentence structures are also provided.
(38) Old Japanese (Taketori: 18)

ito kasikoku tabakari-te Naniha-ni misokani mo-te-ide-nu
very cunningly plan-link Naniha-dat secretly carry-link-come-prf
‘(hei) planned very cunningly, and (hei) came carrying (it) to Naniha secretly’; NOT
‘(hei) planned very cunningly, and (hej) came carrying (it) to Naniha secretly.’
(38′) [(Si) Adv Vi]-te [(Si) (O) PP Adv Vt]

orosokanaruyauni ihi-kere-ba kokoronomamani-mo e-seme-zu
outright speak-prf-link arbitrarily-particle prefix-force-neg
‘(shei) spoke outright, and (shej) could not arbitrarily force (heri) (to marry)’; NOT
‘(shei) spoke outright, and (shei) could not arbitrarily force (herj) (to marry).’
(39′) [(Si) Adv Vi]-ba [(Sj) (Oi) Adv Vt]
Note that in both examples, the argument structure of the first linked unit
is intransitive and that of the second linked unit is transitive, and all
arguments are phonologically unrealized. Hence the only formal clue to
differentiate the working of reference tracking in these examples is the
choice of conjunctive particle. From these considerations, it appears to make
sense to analyse the two conjunctive particles, te and ba, as switch-reference
markers, the former coding the same-subject (SS) relation, and the latter the
different-subject (DS) relation.
However, a closer analysis reveals that te and ba differ in several respects
other than reference tracking, which cannot be explained if we assume that
switch-reference is the defining property of these particles. The primary
evidence for this interpretation comes from the interaction of operator
scope and the layer of linkage. In the RRG conception of sentence structure,
as we have seen in 13.2.1 (10), each layer of linkage has its own operator
specification. What is crucial about te and ba linkages is that they have
different possibilities for the occurrence of operators, in this case the
modal auxiliary.
First, consider the following example involving te-linkage. Here an epi-
stemic auxiliary meru occurs in the final unit of the linkage structure.

550 TOSHIO OHORI

Tubame ko uma-mu-to suru toki-ha, o-wo sasage-te
swallow child bear-aux-comp do time-top tail-acc raise-link
nana tabi meguri-te nan, umi-otosu-meru
seven time go.around-link particle bear-drop-AUX
‘A swallow, when (it) bears a child, may raise its tail and go around seven
times, and then gives birth (to a child).’
In the following example, the modal auxiliary besi, translated here as

‘would’ with an epistemic meaning, also occurs in the final unit of the
linkage structure.

Mosi saihahini kami-no tasuke ara-ba, minami-no umi-ni
if fortunately god-gen help be-link south-gen sea-dat
huka-re-ohasi-nu-besi
blow-pass-go-prf- AUX
‘if fortunately there is God’s help, (you) would be blown and go to the
south sea.’
In (40), meru ‘may’ has scope over all the units linked by te, namely o-wo
sasage ‘raise (its) tail’, nana tabi meguri ‘go around seven times’ and umi-otosu
‘gives birth’. In contrast, in (41) the auxiliary besi has scope only over the
clause where it occurs, namely minami-no umi-ni huka-re-ohasi-nu-besi ‘(you)
would be blown to the south sea’. The difference between the two examples
can be represented schematically as follows, the underlined part indicating
the scope of modal auxiliary:
(40′) te-linkage
[unit 1]-te [unit 2] epistemic modality
(41′) ba-linkage
[unit 1]-ba [unit 2] epistemic modality
This difference in the working of operators is systematic and consistent

in OJ. Given that epistemic modality is characterized as a clause-level
operator, a plausible analysis of these linkage constructions is that (38)
and (40) involving te-linkage are instances of core juncture, while (39) and
(41) involving ba-linkage are instances of clause juncture. That is, the
structure [unit 1] -te [unit 2] together make up a single clause modified
by the epistemic operator. In contrast, for the structure [unit 1]-ba [unit
2], both linked units are clauses and only the latter is within the scope of
the epistemic operator. The implication of this analysis for the issue of
reference tracking is clear. Since core juncture involves argument sharing
as one of its defining features, the retention of subject (more precisely
PSA) in te-linkage is a natural consequence of the typology of
clause linkage.
The analysis of te-linkage as core juncture is further supported by observa-
tions of the behaviour of other operators. In the following example, -mu is

analysed as an auxiliary of volition, and it only modifies the second unit of

linkage. A schematic representation of operator scope is also given.

oni-noyau-naru mono ide-ki-te korosa-mu-to si-ki
devil-like-cop thing appear-come-link kill-AUX -comp do-prf
‘A devil-like monster came up, and (it) meant to kill (me).’
(42′) te-linkage
[unit 1]-te [unit 2] <-root modality
The behaviour of -mu as shown in (42′) is predictable from the RRG theory of
operators and clause linkage because root modality is an operator at the
core level and its scope does not extend to both linked units.
The next question concerning the reference-tracking properties of
clause linkage is why ba-linkage apparently has DS function. Unlike in
the case of te-linkage, where SS function derives straightforwardly from
the nature of clause linkage, the DS function of ba-linkage does not simply
follow from the fact that it is on the clause layer. Here the consideration
becomes largely a pragmatic one. Since ba-linkage is a clause-level con-
struction, it combines separate events. There, the ba-marked clause is
pragmatically detached, serving to provide the background assumption
‘given that/now that’, and this detachment provides a basis for the occur-
rence of a separate actor in the following clause. Also, because of this
detachment, the ba-marked clause assumes a separate viewpoint, which
accounts for the separateness of epistemic stance. Whether this pragmatic
status of the ba-marked clause should be reflected in the structural repre-
sentation, for example as a special kind of PrDP, is an interesting theoret-
ical issue, but at the moment, given the fact that the ba-marked clause is
capable of forming a clause chain, we shall remain rather conservative
about its status and not claim that it exemplifies a new type of
PrDP constituent.
To touch on the semantics of te- and ba-linkage, they both express tem-
poral sequence, but this similarity is only apparent. If we refine our defin-
ition of this notion so it involves temporal proximity, the difference
becomes clear: te-linkage expresses a closely connected sequence, which
often allows a manner or means interpretation, while ba-linkage is not
limited to immediate temporal sequence. This difference in temporal prox-
imity is also accompanied by different manifestation of causality. In te-
linkage, when there is room for causal interpretation, the relation is more
direct and the antecedent event enables the consequent event to happen,
while in ba-linkage, causality holds in the epistemic context, with the
antecedent event providing a reason for the statement given in the conse-
quent event. Here too, the relative strength of clause integration on the
semantic side exhibits correlation with that on the structural side, and this
correlation manifests itself in the difference of reference tracking across
linked units.

552 TOSHIO OHORI
An important piece of evidence for the pragmatically oriented analysis

comes from the following example, where there is an overt PrDP constituent
sentence-initially.

Kono okina-ha, Kaguyahime-no yamome-naru-wo nagekasi-kere-ba,
this old.man-top Kaguyahime-gen single-cop-acc deplore-prf-link
yoki hito-ni aha-se-n-to omohi-hakare-do, . . .
good person-dat marry-caus-aux-comp think-plan-though
‘As for this old man (¼myself ), (I) have deplored Kaguyahime’s being single,
and (I) thought-planned to make (her) marry a good man, but. . .’
This is one of the few exceptions to the generalization that ba-linkage is

associated with DS function in Taketori Monogatari. Here, kono okina ‘this old
man’ refers to the speaker himself and is established as a sentence topic by
the marker ha (¼ modern wa), that is, it occupies the PrDP. The two clauses,
linked by ba, are both pragmatically governed, so to speak, by this topic,
and consequently the PSAs of both clauses, phonologically zero, are
coreferential with it. In other words, the presence of an overt PrDP outside
the linked clauses pre-empts the topic-like interpretation of the ba-marked
clause. When, on the other hand, there is no independent constituent that
occupies the PrDP, the ba-marked clause provides background information
against which another event is reported, with the change of salient
participant.
To summarize, the apparent switch-reference functions of conjunctive
particles in OJ derives from the overall properties of clause linkage, with te
linking cores which share the PSA, while ba links clauses which display a
switch of PSA due to its pragmatic properties.
Beyond Old Japanese, RRG provides many significant insights into
switch-reference phenomena, two of which are briefly reviewed below to
conclude this section. First, the notion of PSA offers a coherent account of
the question of what is monitored in switch-reference. In other theoretical
frameworks, for example Relational Grammar (Farrell et al. 1991), a per-
haps unnecessary theoretical complication arises in order to keep intact
the notion of ‘subject’. In contrast, the notion of PSA in RRG is itself a
relativized one consisting of a hierarchically arranged array of macroroles,
and the theory of reference tracking needs only to make reference to
PSA, without any additional stipulation as to what is monitored for
referential identity.
Second, RRG adopts a parallel architecture approach to grammar, and
hence no single subsystem dominates the rest. In contrast, in Generative
Grammar, which is exclusively syntax-driven, switch-reference must be
handled in terms of tree configuration, whether by government (as in Finer
1985) or by some other mechanism such as agreement. But this way of
treating switch-reference has been seriously questioned by linguists working
in other frameworks (for example, Roberts 1988; Ohori 2005). In RRG, the

layered representation of the sentence structure in principle provides con-

straints on, not determines, the interpretation of coreference relations.
Reference tracking is computed through the interaction of linkage types,
the properties of PSA, and the semantic/pragmatic information of switch-
reference morphemes, as has been demonstrated in this section.
13.5 Concluding Remarks
In this chapter, we have surveyed the RRG approach to complex structures,

known as the theory of clause linkage. More than a generation has passed
since the two parameters and values for the characterization of complex
structures, namely juncture and nexus, were originally proposed, but they
remain intact, supported by rich empirical data and insights therefrom.
This chapter has also reviewed more recent innovations in the theory, such
as the incorporation of the constructs such as pre-core slot and pre-
detached position as well as refinement of the list of semantic relations
between the linked units. It still remains to be seen in which way the
recent ‘quantitative turn’ in theoretical linguistics will affect our concep-
tion of complex structures (cf. Bickel 2010), but the fundamental notions
behind the RRG typology which have been expounded in this chapter will
remain crucial ingredients of any attempt to systematically define clause
linkage typology.
References
Akiba, Katsue. 1977. Switch-reference in Old Japanese. In Kenneth Whistler,

Robert D. Van Valin, Jr., Chris Chiarello, Jeri J. Jaeger, Miriam Petruck,
Henry Thompson, Ronya Javkin and Anthony Woodbury (eds.), Proceedings
of the Third Annual Meeting of the Berkeley Linguistics Society, 610–619. Berke-
ley, CA: Berkeley Linguistics Society.
Akiba, Katsue. 1978. A Historical Study of Old Japanese Syntax. PhD dissertation,
University of California at Los Angeles.
Bauer, Winifred. 1993. Maori. London: Routledge.
Bickel, Balthasar. 1993. Belhare subordination and the theory of topic. In
Karen H. Ebert (ed.), Studies in Clause Linkage. Papers from the First Köln-Zürich
Workshop, 23–55. Zurich: Seminar für Allgemeine Sprachwissenschaft.
age: a multivariate analysis. In Bril (ed.), 51–101.
Bril, Isabelle (ed.). 2010. Clause Linking and Clause Hierarchy: Syntax and Prag-
matics. Amsterdam: John Benjamins.
Burton-Roberts, Noel. 2005. Parentheticals. In Keith Brown (ed. in chief ),
Encyclopedia of Language and Linguistics, Vol. 9 (2nd ed.), 179–182. Amster-
dam: Elsevier.

554 TOSHIO OHORI
Croft, William. 2001. Radical Construction Grammar: Syntactic Theory in Typo-

logical Perspective. Oxford: Oxford University Press.
Dedrick, John M. and Eugene H. Casad. 1999. Sonora Yaqui Language Structures.
Tucson, AZ: University of Arizona Press.
Diller, A. V. N. 2006. Thai serial verbs: Cohesion and culture. In Alexandra Y.
Aikhenvald and R. M. W. Dixon (eds.), Serial Verb Constructions: A Cross-
Linguistic Typology, 160–177. Oxford: Oxford University Press.
Donohue, Mark. 1999. A Grammar of Tukang Besi. Berlin: Mouton de Gruyter.
Farrell, Patrick, Stephen A. Marlett and David M. Perlmutter. 1991. Notion of
subjecthood and switch reference: evidence from Seri. Linguistic Inquiry 22:
431–456.
Fillmore, Charles J., Paul Kay and Mary Catherine O’Connor. 1988. Regularity
and idiomaticity in grammatical constructions: The case of let alone. Lan-
guage 64: 501–538.
Finer, Daniel L. 1985. The syntax of switch-reference. Linguistic Inquiry 16:
35–55.
Foley, William. 2010. Clause linkage and nexus in Papuan languages. In Bril
(ed.), 27–50.
Foley, William and Robert D. Van Valin, Jr. 1984. Functional Syntax and
Givón, Talmy. 1980. The binding hierarchy and the typology of comple-
ments. Studies in Language 4: 333–377.
Givón, Talmy. 1985. Iconicity, isomorphism, and non-arbitrary coding in
syntax. In John Haiman (ed.), Iconicity in Syntax, 187–219. Amsterdam: John
Benjamins.
Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to
Argument Structure. Chicago, IL: University of Chicago Press.
Haiman, John. 1985. Natural Syntax: Iconicity and Erosion. Cambridge: Cam-
Haiman, John and Sandra A. Thompson. 1984. ‘Subordination’ in universal
grammar. In Claudia Brugman and Monica Macaulay (eds.), Proceedings of
the Tenth Annual Meeting of the Berkeley Linguistics Society, 510–523 Berkeley,
CA: Berkeley Linguistics Society.
Hasegawa, Yoko. 1996. A Study of Japanese Clause Linkage: The Connective TE in
Japanese. Stanford: CSLI Publications/Tokyo: Kurosio.
Hengeveld, Kees and J. Lachlan Mackenzie. 2008. Functional-Discourse Gram-
mar: A Typologically-Based Theory of Language Structure. Oxford: Oxford Uni-
versity Press.
Kaltenböck, Gunther, Bernd Heine and Tania Kuteva. 2011. On thetical
grammar. Studies in Language 35: 848–893.
Lehmann, Christian. 1988. Towards a typology of clause linkage. In John
Haiman and Sandra A. Thompson (eds.), Clause Combining in Grammar and
Discourse, 181–225. Amsterdam: John Benjamins.
McAuley, T. E. 2002. Switch-reference and semantic discontinuity in Late Old
Japanese. Journal of Japanese Linguistics 18: 29–58.

Mous, Marten. 1993. A Grammar of Iraqw. Hamburg: Buske.

Nikolaeva, Irina and Maria Tolskaya. 2001. A Grammar of Udihe. Berlin:
Mouton de Gruyter.
Ohori, Toshio. 1992. Diachrony in Clause Linkage and Related Issues. PhD
dissertation, University of California at Berkeley.
Ohori, Toshio. 1994. Diachrony of clause linkage: TE and BA in Old through
Middle Japanese. In William Paliugca (ed.), Perspectives on Grammaticaliza-
tion, 135–149. Amsterdam: John Benjamins.
Ohori, Toshio. 2001. Some thoughts on a new systematization of interclausal
semantic relations. Paper presented at the Role and Reference Grammar
Workshop, Santa Barbara.
Ohori, Toshio. 2002. Kootai-shiji koo-bun no tsuujisoo: toogo-henka to
kategorii-ka (Diachrony of switch-reference constructions: syntactic
change and categorization). In Toshio Ohori (ed.), Ninchi Gengogaku II:
Kategorii-ka (Cognitive Linguistics II: Categorization), 297–321. Tokyo: Uni-
versity of Tokyo Press.
Ohori, Toshio. 2005. Construction Grammar as a conceptual framework for
linguistic typology: A case from reference tracking. In Mirjam Fried and
Hans C. Boas (eds.), Grammatical Constructions: Back to the Roots, 215–237.
Olson, Michael L. 1981. Barai Clause Junctures: Toward a Functional
Theory of Interclausal Relations. PhD dissertation, Australian National
University.
Pavey, Emma L. 2010. The Structure of Language: An Introduction to Grammatical
Roberts, John R. 1988. Amele switch-reference and the theory of grammar.
Linguistic Inquiry 19: 45–64.
Shibatani, Masayoshi. 1972. Three reasons for not deriving ‘kill’ from ‘cause
to die’ in Japanese. In John P. Kimball (ed.), Syntax and Semantics 1, 125–137.
Silverstein, Michael. 1993. Of nominatives and datives: Universal Grammar
from the bottom up. In Van Valin (ed.), 465–498 (orig. unpublished
ms. 1981).
Van Valin, Robert D., Jr. 2021. Cosubordination. In Robert D. Van Valin, Jr.
(ed.), Challenges at the Syntax–Semantics–Pragmatics Interface. Newcastle upon

556 TOSHIO OHORI
Vásquez Soto, Verónica. 2002. Some constraints on Cora causative construc-

tions. In Masayoshi Shibatani (ed.), The Grammar of Causation and Interper-
sonal Manipulation, 197–244. Amsterdam: John Benjamins.
Watters, James K. 1993. An investigation of Turkish clause linkage. In Van
Valin (ed.), 535–560.
Weber, David John. 1983. Relativization and Nominalized Clauses in Huallaga
(Huánuco) Quechua. Berkeley, CA: University of California Press.
Weber, David John. 1989. A Grammar of Huallaga (Huánuco) Quechua. Berkeley,
CA: University of California Press.

14
Linking Syntax and
Semantics in Adverbial
(Adjoined) Clauses
Lilián Guerrero
Abbreviations

AFFIRM affirmative M male speaker

ASF adjective suffix O object
CLM clause linkage marker P person
CO coreferential POT potential
DIR directional RED reduplication
DM discourse marker RM remote
DP dependent marker SEQ sequential
EMPH emphatic STV stative
IF illocutionary force UNR unrealized
INT intensifier UNSP.O unspecified object
NPF noun prefix
14.1 Introduction
One of the significant features of Role and Reference Grammar (RRG) is that
it seeks to explain the universal features of clauses without imposing fea-
tures on languages in which there is no evidence for them.* A second
distinctive aspect is that it provides the necessary mechanisms for describ-
ing linguistic phenomena with scopes not only in simple clauses but also in
complex constructions. A third defining feature is the theory of clause
linkage. In RRG, the analysis of complex constructions is grounded in three
* This study was partially supported by CONACyT-Ciencia Básica (A1-S-24378) and the Ministerio de Ciencia, Innovación y
Universidades, Gobierno de España (FFI2017–85429-P) grants. I’m grateful to the anonymous reviewers and the editors of
this Handbook for their comments and suggestions on the first version of the chapter.

558 LILIÁN GUERRERO
components: the level of juncture (nuclear, core, clausal and sentence), the
nexus relations (coordination, subordination and cosubordination), and the
semantic relationships between the two units. Unlike the traditional
coordination–subordination dichotomy, this model recognizes eleven
nexus–juncture combinations. The theory of clause linkage in RRG is crucial
to the analysis of adverbial (adjoined) clauses.
Typically, adverbial clauses are a type of subordination. When compared
with relative and complement subordinate clauses, however, adverbial
clauses are considered the least subordinate type (Matthiessen and Thomp-
son 1988; Givón 2001; Thompson et al. 2007). On one hand, adverbial
clauses do not fill an obligatory slot in the main clause and there is no
evidence that they are syntactically embedded (Diessel 2013); hence, adver-
bial clauses are usually considered adjuncts of the main unit in the periph-
ery or a peripheral position. On the other hand, they express several
semantic meanings including location, time, manner, purpose, condition,
reason, and so on; these meanings do not depend on the main verb or
clause. In fact, adverbial clauses subsume a wide range of structures with
varying syntactic properties that often overlap with the corresponding
features of coordinated constructions and other types of subordinated
sentences (Lehmann 1988; Cristofaro 2003; Hetterle 2015). For instance,
some but not all adverbial clauses are introduced by a subordinator or
clause-linkage marker (CLM); compare (1a) with (1b–d). Some CLMs have
lexical meaning (e.g. if in (1d)), while others do not (e.g. to in (1b)). Some
adverbial clauses show a flexible order with respect to the main clause (e.g.
temporal clauses (1c)), but others may prefer a fixed position (e.g. condition-
als (1d)). In addition, some adverbial clauses make use of tighter syntactic
linkages, while others are encoded by looser syntactic combinations.
A piece of evidence for a tighter syntactic linkage comes from the coding
of shared arguments: in (1a–b), the two units share the privileged syntactic
argument (PSA, i.e. traditional subject) and so it is coded only once in the
construction; in (1c), the PSAs are also the same, but each unit codes its own
argument. The marking of tense-aspect-mood (TAM) operators is also rele-
vant; in (1a–b), the linked verb depends on TAM operators from the main
verb, while in (1c–d), the linked verb is completely independent of TAM
marking and can be independently negated.
(1) English
a. Kim left yelling.
b. John went to the store to buy milk.
c. When he left the office, he forgot to turn the lights off.
d. If it rains, I won’t go to the movies.
The main goal of this chapter is to examine the set of rules that relates
syntactic and semantic representations to each other in adverbial (adjoined)
clauses.1 The analysis is based on argument sharing, operator sharing, and
the presence or absence of CLM, as well as the semantic cohesion between

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses 559
the two units. In order to show the structural diversity of adverbial clauses,
the analysis is based on cross-linguistic data, though special attention is
given to English, Spanish and Yaqui, the languages I am most familiar with.
The content is organized as follows. Section 14.2 briefly introduces the RRG
theory of clause linkage, while Section 14.3 lays out the linking algorithm.
Section 14.4 examines the syntax and semantics of adverbial clauses; the
discussion is organized in five groups depending on the adverbial relation:
concessive and conditional clauses (14.4.1), reason clauses (14.4.2), temporal
clauses (14.4.3), purpose clauses (14.4.4), and manner and means clauses
(14.4.5). Section 14.5 summarizes the merits of RRG in capturing the com-
plex patterns of adverbial clauses.
14.2 The Theory of Clause Linkage in RRG
RRG proposes three main components for the study of clause linkage: the
theory of juncture, the theory of nexus, and the interclausal semantic
relation. The first two parameters concern the syntactic representation of
complex constructions, while the last refers to their semantic representa-
tion. The theory of juncture deals with the units being linked. In a
nuclear juncture, there is a single core containing two nuclei; verbs in
a nuclear juncture take a single set of core arguments, that is, they form
a single complex predicate. In a core juncture, there is a single clause
containing more than one core, each with its own set of arguments; the
linked units can share part of the argument structure of the verbs. In a
clausal juncture, whole clauses are joined, and each clause may be fully
independent of the others. The schematic representation and examples
of juncture types are shown in (2). The examples come from Van Valin
(2005: 198).
(2) a. [CORE… [ NUC… ] … þ… [ NUC… ] … ] Nuclear juncture

a′. Vince has wiped the table clean.
b. [ CLAUSE… [ CORE… ] … ] … þ … [ CORE… ] … ] Core juncture
b′. Fred saw Harry leave the room.
c. [ SENTENCE… [ CLAUSE… ] … þ … [ CLAUSE… ] … ] Clausal juncture
c′. Kim cried because Leslie didn’t call.
The theory of nexus concerns the syntactic relationship between the

units in the juncture, termed subordination, coordination and cosubordi-
nation, which are distinguished on the basis of structural and operator
dependency. In coordination, the two units are added together in a
sequence in a relationship of equivalence and independence. (3a) is an
example of coordination taking a conjunction, but many languages have
paratactic coordination lacking CLMs. In subordination, the linked unit
functions either as an argument (daughter subordination) or as a modifier
(ad-subordination). The dependent clauses in (3b–d) function as core

arguments of the main verb, whereas the linked unit in (3e) is an adjunct
modifier of the main clause. Nuclear, core and clausal units modifying the
main clause are referred to as ad-nucleus, ad-core and ad-clausal subordin-
ation. The examples in (3), and part of the discussion that follows, come
from Matić et al. (2014: 5–9).2
(3) English coordination and subordination

a. The tall man went to the party and the woman went shopping.
b. Sam believed that Bill stole the money.
c. Mary regretted kissing Bill the most.
d. Sami tried _i to open the door.
e. John confronted Bill after the police arrived.
The typical example of daughter subordination is the use of a unit as a

core argument of a matrix predicate, and it has the distributional properties
of simple referential phrases (RPs). When functioning as the direct object
(3b), this unit can become the PSA of a passive construction (that Bill stole the
money was believed by Sam) and can be it-clefted (it was that Bill stole the money
that Sam believed). Gerund complements also have this property in English
(e.g. kissing Bill was regretted by Mary the most, it was kissing Bill that Mary
regretted the most for (3c)). In contrast, infinitival complements cannot
become the passive-PSA or be it-clefted (e.g. *to open the door was tried by
Sam, *it was to open the door that Sam tried for (3d)). Both gerund and infinitival
complements are subjectless, non-finite and sub-clausal linkage units, but
the gerund shows the same properties as a simple RP argument, while the
infinitive does not. That is, the infinitive in (3d) cannot be analysed as
daughter subordination. This clause linkage cannot be analysed as an
instance of ad-subordination either. Ad-subordinate units are canonically
modifiers, and they are marked by an explicit CLM, usually an adposition, as
after in (3e), hence the adverbial unit is often embedded as the object of that
adposition. Another typical feature of ad-subordination is that it can occur
either before or after the clause it modifies (e.g. after the police arrived, John
confronted Bill for (3e)). The infinitival complement does not modify the main
clause, there is no explicit CLM, and it cannot occur clause initially (e.g. *to
open the door Sam tried). Hence, infinitival complements do not correspond to
ad-subordination.
The infinitival complement in (3d) can be considered an instance of
coordination or cosubordination. Cosubordination is defined as ‘a kind of
dependent coordination, in which units of equivalent size are joined
together in a coordinate-like relation but share some grammatical category’
(Van Valin 2005: 187). A crucial property of this nexus relation is operator
sharing. In cosubordination the linked verb is dependent upon the matrix
verb for expression of one or more operators at the level of juncture,
whereas in coordination, the two verbs can be, but are not necessarily,
independently specified for the relevant operators. In (3d), the linked unit
depends on TAM marking from the main verb, and it cannot be independ-
ently negated (e.g. *Sam tried not to open the door); therefore (3d) is an instance

of cosubordination. However, the infinitival complement in (2b) can be

independently negated (e.g. Fred saw Harry not leave the room); therefore (2b)
is an example of core coordination.
It is important to keep in mind that these juncture–nexus relations are
abstract syntactic relationships that can correspond to more than one
formal construction type. For instance, the infinitival complement in (2b)
corresponds to core coordination, while the one in (3d) yields core cosubor-
dination. The -ing constructions in (4) provide another example of the lack of
one-to-one mapping between juncture–nexus and a formal construction
type. The linked unit serves as a nominal argument in (4b, c), and as an
adverbial clause in (4a, d–f). In the last two examples, the linked unit has
even acquired some nominal features: either the subject takes the form of
an accusative/genitive pronoun (4e) or there is a predicative preposition
taking a core as an object (4f). Examples (4a–d) come from Van Valin (2005:
197), and those in (4e–f) from Pérez Quintero (2002: 41).
(4) English gerund clauses

a. Kim sat reading a book. Core cosubordination
b. Dana saw Chris washing the car. Core coordination
c. Washing the car today would be a mistake. Core subordination (daughter)
d. Pat ran down the hall laughing loudly. Clausal cosubordination
e. Them trying to sing a song was just too horrible. Core subordination (daughter)
f. On arriving in Oxford, the first thing he did was Core subordination (adjoined)
to visit his old supervisor.
The coding of identical arguments is another crucial feature in determin-

ing particular syntactic combinations; they are treated as coreferential in
clausal junctures, and as shared arguments in core junctures. In (4a), the
main agent obligatorily controls the identity of the missing argument. (4b) is
understood as Dana saw Chris while Chris was washing the car, in which case the
stimulus of ‘see’ controls the identity of the agent of ‘wash’.3 Core
subordination does not impose any reference constraint and all the argu-
ments of the dependent unit can be expressed by either RPs or pronouns, as
in (4e).
In addition to the nine juncture–nexus types that result from the three
primary linked units (nuclear, core and clause) and the three nexus types
(cosubordination, coordination and subordination), there are two other
syntactic combinations: sentential subordination, as in After Anna finished
her work, she went to the party, and sentential coordination, like in As for Tom,
Mary talked to him, but as for Sam, she refused. The lack of operators at the level
of the sentence rules out sentential cosubordination. A language does not
need to have all eleven juncture–nexus types, and in fact most do not;
English, for instance, exhibits nine combinations (Van Valin 2005: 197).
There is a basic principle governing clause linkage: the unmarked linkage
involves units at the same level of juncture: nucleus with nucleus, core with
core, clause with clause, and sentence with sentence. The typical exception
to this is complementation in which there is a clause functioning as a core

argument. Moreover, the type of juncture is determined by the level at

which the linkage occurs instead of the size of the linked unit. Compare
the sentence that she arrived late shocked everyone, and the alternative structure
it shocked everyone that she arrived late. In both cases the linked unit is a clause,
but they are linked at different levels. In the former, the embedded clause
functions as a core argument, whereas in the latter it is the pronoun it that
occupies that syntactic slot. The linked clause (which expresses the content
of it) is outside the core and a direct daughter of the higher clause node.
Then, there is a clausal juncture and a core juncture, respectively. The first
case is a good example of an asymmetrical linkage since the linked unit is
contained within a sub-clausal unit.
The selection and form of the CLM has been a crucial topic in the study of
adverbial clauses because they can specify the semantic relation between
the main clause and the adverbial clause. There are two types of adverbial
markers (Thompson et al. 2007): grammatical morphemes with no lexical
meaning (e.g. English to (1b)), and grammatical morphemes with lexical
meaning (e.g. English if (1d)). Some languages have a large inventory of
adverbial CLMs, while others have few markers. Barbareño (Chumash, Cali-
fornia) is an example of the first group (Wash 2001). The adverbial conjunc-
tions maĺi ‘when’, na ‘if/when’, wa ‘if/when’ all code temporal meanings (5a),
as well as other kinds of semantic relations (5b) (i.e. semantically under-
specified). In contrast, Ɂakimpi ‘during, while’ (5c), and kayké ‘because’ (5d)
introduce simultaneous and reason relations, respectively (i.e. semantically
specified).
(5) Barbareño (Chumash; Wash 2001: 146–168, 309)

a. [na¼mik ni¼p-ušhoɁ Ɂi¼s-xiliwokoy [maĺi¼k-saɁ-aqmil]
if¼long dp¼2-leave top¼3-settle when-1-fut-drink
‘If/when you leave it long, it settles before I drink it.’
b. [maĺi moke ka¼s-exšwe hi¼hoɁ¼Ỉ¼sewu]
when already then¼3-melt dp¼dist¼art¼fat ˘
hika¼s-am-apšik hi¼l¼čtaniw hı¼l¼woti
seq¼3¼indf-pour.into dp¼art¼small dp¼art¼can
‘When/since the oil was melted, they poured it off into a small can.’
c. [Ɂakimpi hu¼how̉ o Ɂal-tupmekč hi¼luwisa]
during rm¼still stv-child dp¼Luisa
Ɂi¼s-am-monus-waš hi¼hoɁ¼s-Ɂasas
top¼3-indf-paint-pst dp¼dis¼3-chin
‘While Luisa was a girl, they tattooed her chin.’
d. s-am-sumoč-wún hi¼noɁ¼l¼en-Ɂenxweq-Ɂ
3-indf-tattoo-pl.o dp¼dis¼art¼adolescent.girl-emph
[kayké Ɂalka¼Ɂišnaniš- waš]
because stv.then¼custom- pst
‘They tattooed the girls because it was the custom.’
All juncture–nexus types are used to express certain semantic relations

between the units in the juncture (e.g. causation, purpose, perception,

temporal, and so on). Importantly, a given semantic relation may take more
than one juncture–nexus type. For instance, in Kokama (Tupian, Peruvian
Amazon), there are three purposive constructions that exhibit different
degrees of integration with the main unit. In the examples in (6), the linked
verb is not marked by TAM values, and the adverbial unit cannot take the
negative particles that operate in simple clauses (operator dependency); they
differ in terms of coreferentiality conditions (Vallejos 2014: 40). In the first
two clauses, there is a missing argument which must be coreferential with an
argument within the main clause; in tara constructions, the main absolutive
argument controls the obligatorily omitted nominative PSA (e.g. in order for
Magdalena to cure Jose), whereas in mira constructions, the main absolutive
controls the obligatorily omitted accusative argument in the linked unit (e.g.
in order for Jose to cure Magdalena). In tsen constructions, there are no
reference constraints, and all the purposive arguments must be expressed
by either RPs or pronouns; in fact, ra ‘3sg.m’ can refer to new entities (for
example, in order for someone else to cure Jose). Most likely, (6a–b) involve
clausal cosubordination, and (6c) clausal subordination.
(6) Kokama (Tupian; Vallejos 2014: 40)
a. Mijiri erura Mararina-uyi [Kutsi(¼pura) _i mutsanaka-tara]
Miguel bring Magdalena-pst1 Jose¼foc cure-clm
‘Miguel brought Magdalena in order for her to cure Jose.’
b. Mijiri erura Mararina-uyi [_i Kutsi(¼pura) mutsanaka-mira]
Miguel bring Magdalena- pst1 Jose¼foc cure-clm
‘Miguel brought Magdalena in order for Jose to cure (her).’
c. Mijiri erura Mararina-uy [Kutsi(*¼pura) ra mutsanaka-tsen]
Miguel bring Magdalena- pst1 Jose¼foc 3sg.m cure-clm
‘Miguel brought Magdalena in order for her/another to cure Jose.’
In the Kokama examples, syntactic integration correlates with semantic–

discourse integration (Vallejos 2014: 41). In addition to referential control,
the interpretation in (6a–b) is that the event in the linked unit occurs right
after the main event, while in (6c) this temporal integration does not exist.
In (6a–b), the purpose event is successfully realized, and the main event is a
condition for the purposive event to take place, but in (6c) it is not.
The third component of the RRG theory of clause linkage is the inter-
clausal semantic relations. The semantic integration between the two
units is the result of combining a number of more basic semantic notions
including, but not limited to, the five hierarchies given in (7) (Van Valin
2005: 211; Guerrero 2009: 328, 2013: 17). In particular, the hierarchy in (7e),
proposed in Guerrero (2013), predicts that the referential identity of an
obligatory controlled argument is determined first by the main actor or
undergoer, and then by another main core argument. In RRG terms, the
above tara and mira purposive constructions in Kokama show a closer
integration based on all the hierarchies, whereas the tsen construction
yields a less close integration since it is temporally unspecified and does
not demand participant sharing.

(7) a. Temporal hierarchy:

phase of a single event > simultaneous events > sequential events >
unspecified
b. Causal hierarchy:
physical > verbal > underspecified[non-defeasible] > inferred[defeasible]
c. Participant’s mental disposition:
Intention > internal/direct experience > mental experience: commitment >
mental experience: reasoning > non-mental experience: report
d. Necessarily shared participant [NSP]:
Yes > No
e. Referential control hierarchy:
main actor > main undergoer > another main core argument > there is
no control relation
The juncture–nexus relations and the semantic interclausal relations

form the Interclausal Relations Hierarchy (Figure 14.1). The syntactic com-
binations are ranked in terms of syntactic tightness. The linkage types at
the bottom are combinations of whole clauses constituting sentences (e.g.
Tyron talked to Tanisha, and Yolanda chatted with Kareem), where two inde-
pendent units are temporally unordered. As one goes up the hierarchy, the
linked units lose more and more features of an independent clause until
they are reduced to a bare nucleus or predicate (e.g. nuclear co-
subordination, as in Max seemed tired). That is, linkage types at the bottom
Strongest Closest
Nuclear co-subordination Causatives [1]
Phase
Nuclear subordination Manner
daughter Motion
peripheral Position
Nuclear coordination Means
Psych-action
Core co-subordination Purposive
Jussive
Core subordination Causatives [2]
daughter
peripheral
Direct perception
Core coordination Indirect perception
Propositional attitude
Clause co-subordination Cognition
Indirect discourse
Clause subordination Direct discourse
daughter Circumstances
peripheral Reason
Clause coordination Conditionals
Concessive
Sentential subordination Simultaneous actions
Sequential actions
Sentential coordination Situation-situation: unspecified
Weakest Loosest
Syntactic relations Semantic relations
Figure 14.1 Interclausal Relations Hierarchy (Van Valin 2005: 209)

are less tight than linkage types at the top end. In the middle portion, in a
non-subordinate core juncture, the two cores obligatorily share one core
argument; in a subordinate core juncture, the linked core serves as a
syntactic argument of the matrix core. The semantic relations are also
organized according to the degree of semantic cohesion between the two
units, that is, the degree to which they express facets of a single action or
discrete actions or events.
There is an iconic principle governing the interaction of the two hierarch-
ies: the closer the semantic relation between two propositions, the stronger
the syntactic link joining them. That is, the semantic relations at the top
end of the hierarchy should be realized by the linkage categories at the top
as well, and the relations at the bottom of the hierarchy should be realized
by the linkage categories at the bottom of the syntactic side. Moreover, while
there is often more than one syntactic realization of a particular semantic
relation, the tightest syntactic linkage realizing it should be tighter than the
tightest syntactic linkage realizing looser semantic relations. The syntactic
relations and the semantic relations are linked together following a set of
mechanisms or linking algorithm.
14.3 The Linking Algorithm
Within RRG, the algorithm system links the syntactic, semantic and prag-
matic components in a sentence, and it includes universal as well as
language-specific components (Van Valin 2005: 129). The system is bidirec-
tional, that is, it links the semantic representation to the syntactic repre-
sentation (language production), and the syntactic representation to the
semantic representation (language comprehension). The algorithm for
linking semantics to syntax for simple sentences is summarized in (8); see
also Chapter 1 of this handbook.4
(8) Linking algorithm: semantics ! syntax (Van Valin 2005: 225–226)

1. Construct the semantic representation of the sentence, based on the logical
structure of the predicator.
2. Determine the actor and undergoer assignment following the Actor-
Undergoer Hierarchy
3. Determine the morphosyntactic coding of the arguments
a. Select the privileged syntactic argument (PSA), based on the PSA
Selection Hierarchy (a′) and the relevant principles of accessibility (a′′)
a′. PSA Selection Hierarchy: Arg of DO > 1st arg of do′ >1st arg of pred′
(x, y) > 2nd arg of pred′ (x, y) > arg of pred′ (x)
a′′. Accessibility to PSA principles
a. Accusative constructions: highest-ranking direct core argument in (a′)
(default)
b. Ergative constructions: lowest-ranking direct core argument in (a′)
(default)
c. Restrictions on PSA in terms of macrorole status:

1. Languages in which only macrorole argument can be PSA (e.g.

German, Italian, Dyirbal, Jakaltek, Sama, Yaqui)
2. Languages in which non-macrorole arguments can be PSA (e.g.
Icelandic, Georgina, Japanese, Korean, Kinyarwanda)
d. Restrictions in terms of argument coding:
1. Languages with case-sensitive PSAs (e.g. English, German, Nepali,
Maithili)
2. Languages with case-insensitive PSAs (e.g. Belhare, Tibetan)
b. Assign the arguments the appropriate case markers and/or adpositions
following the case-marking rules and the postpositional rules for the
language.
c. Assign the agreement marking to the main or auxiliary verb, as appropriate.
4. Select the syntactic template(s) for the sentences:
a. Syntactic template selection principle: The number of syntactic slots for
arguments and argument-adjuncts within the core is equal to the
number of distinct specified argument position in the semantic
b. Language-specific qualifications of the principle in (a):
2. Argument-modulation voice constructions reduce the number of
core slots by 1.
3. The occurrence of a syntactic argument in the pre-/post-core slot
reduces the number of core slots by 1 (may override b1).
5. Assign arguments to positions in the syntactic representation of the sentence.
a. Assign the [WH] argument(s) to the appropriate position(s) in the clause.
b. If there is a [þWH] argument of a logical structure:
1. assign it to the normal position of a non-wh-argument with the same
function, or
2. assign it to the pre-core slot or post-core slot, or
c. A non-wh-argument may be assigned to the pre-core or post-core slot,
subject to focus structure restrictions (optional).
d. Assign the [WH] argument(s) of the logical structure(s) other than that
of the predicator in the nucleus to
1. a periphery (default)
a. if the representation is pred′ (RP/LS, LSmain ) where pred′ is an
adpositional predicate, then assign the P þ RP/Core/Clausal to the
peripherycore .
b. if the representation is LSmain pred′/connective′ RP/LS, then
assign the P þ RP or CLM þ Clause to the peripheryclause .
3. the left- or right-detached position.
The algorithm for linking syntax-to-semantics for simple sentences is

presented in (9).
(9) Linking algorithm: syntax ! semantics (Van Valin 2005: 226–227)

1. Determine the macrorole(s) and the other core argument(s) in the clause
based on the logical structure (LS) of the predicator.

a. If the verb is intransitive, then assign the PSA either to the macrorole
or direct core argument status, depending upon the language
(language-specific)
b. If the verb is transitive and the language lacks voice oppositions,
determine the macroroles from case marking and/or word order
(language-specific)
c. If the language has a voice opposition, determine the voice of a
transitive verb (language-specific)
1. If the construction is syntactically accusative:
a. if it is the unmarked voice, the PSA is the actor
b. if it is passive, the PSA is not the actor of the predicate in the nucleus
1. the actor may appear as a direct core argument (language-
specific); or
2. the actor may appear in the peripherycore marked by an
adposition or oblique case (language-specific); or
3. if there is no actor in the core or the periphery, then replace
the variable representing the highest ranking argument in
the LS with ‘Ø’.
2. If the construction is syntactically ergative:
a. if it is the unmarked voice, the privileged syntactic argument is
undergoer.
b. if it is antipassive, the privileged syntactic argument is actor:
1. the undergoer may appear as a direct core argument or as an
oblique element (language-specific);
2. if there is no undergoer in the core or the peripherycore ,
then replace the variable representing the lowest ranking
argument in the LS with ‘Ø’.
3. Assign macrorole status to the other direct core argument, if it is
not dative or in an oblique case (language-specific).
d. If the language is head-marking and there are independent RPs in the
clause, associate each RP with a bound argument marker (language-
specific).
2. Retrieve from the lexicon the LS of the predicate in the nucleus of the
clause and execute step 2 from (8) with respect to it, subject to the
following provisos:
a. If the language allows variable undergoer selection and if there is more
than one choice for undergoer, do not assign undergoer to an
argument in the LS.
b. Determine the linking of the non-macrorole core argument:
1. If there is a two-place state predicate in the LS and if the non-
macrorole core argument is marked by a locative adposition or
dative or a locative-type case, then link it with the first argument
position in the state predicate in the LS and link the other non-actor
core argument (if there is one) to the second argument position in
the state predicate, or
2. If there is a two-place state predicate in the LS and if the non-macrorole
core argument is not marked by a locative adposition or dative or a
locative-type case, then link it with the second argument position in the
state predicate and link the other non-actor core argument (if there is
one) to the first argument position in the state predicate.

3. Otherwise, link the animate RP with the first argument position

in the state predicate in the LS.
3. Link the arguments determined in step 1 with the arguments
determined in step 2 until all core arguments are linked.
4. If there is a predicative adpositional adjunct, then retrieve its LS from
the lexicon, insert the LS of the core as the second argument in the LS
and the object of the adposition as the first argument.
5. If there is an element in the pre- or post-core slot (language-specific):
a. Assign it the remaining unlinked argument position in the semantic
b. And if there are no unlinked argument positions in the sentence,
then treat the wh-word like a predicative preposition and follow the
procedure in step 4, linking the wh-word to the first argument
position in the LS.
Most complex sentences follow the linking system developed for simple
clauses. For instance, the units involving clausal junctures behave largely
like independent clauses. Consider the following English constructions (Van
Valin 2005: 228–229).
(10) a. Dana jogged through the park, and Kim waved to him
b. Kim worked on the assignment in the morning and __i/*j will finish it in
the afternoon
In the conjunction construction in (10a), there are two unordered states

of affairs, each with its own argument structure. Each unit is linked
independently of the other, as if each unit were a simple sentence on its
own, and it follows these steps from the semantics-to-syntax linking in (8):
construct the semantic representation (step 1), assign the actor and the
undergoer (step 2), and determine the morphosyntactic coding of the
arguments for each unit (step 3). After selecting the syntactic templates
for each unit (step 4), select a sentence template that involves two coordin-
ated units; the conjunction is attached to the second clausal template.
Finally, assign arguments to positions in the syntactic representation of
the sentence for each unit; the peripheral elements are linked last, after all
of the core-internal arguments are dealt with (step 5). This is a typical case
of clausal coordination since the two units are independent of each other
in terms of TAM values and illocutionary force (IF). The construction in
(10b) is structurally very similar to (10a), except that there is a missing
argument in the second unit (co-indexed with the main actor Kim) and the
IF must be shared across all conjuncts. A sentence like *Did Kim work on the
assignment in the morning and will finish it in the afternoon? is ruled out
because only the first conjunct is questioned. This yields clausal cosubor-
dination.5 In English, only highly topical elements can receive zero coding,
and the PSAs in (10b), which are pragmatic controllers and pivots, are
topics (Van Valin 2015: 229). As in (10a), the clauses in (10b) link separately

of each other, but the construction as a whole imposes a coreferential

constraint when linking the non-initial conjunct.
In Yaqui, coreferential PSAs in conjunction constructions marked by into
‘and’ (11a) can be omitted or be expressed by a full (inepo ‘I’) or reduced
nominative pronoun (ne ‘I’) in the second component. However, topical
elements usually receive zero marking. In the fragment in (11b), the con-
troller ume chapayecam ‘the Chapayecas’ is an argument of the initial clause.
There is a missing argument in the five non-initial clauses co-indexed to the
pragmatic controller. Note that there is no CLM linking the non-initial
clauses, and only the last verb takes TAM operators. Then, conjunction
constructions in Yaqui yield clausal cosubordination, because the linked
units show operator dependency as well as argument dependency.
(11) Yaqui
a. Inepo teopo-u bwite-k into (ne) Peo-ta bicha-k
1sg.nom church-dir run.sg.pfv clm 1sg.nom Peter-acc see-pfv
‘I ran to the church, and then (I) saw Peter.’
b. [U-me chapayeca-mi yi’i-su-k-o], _i maska-ta emo u’ura,
the-pl chapayeca-pl dance-cmpl-pfv-clm mask-acc self take.off
teopo bicha-po _i im am¼jajawa, emo koba-t _i
church towards-loc here 3pl.acc¼red.leave self head-loc
payum-mea papatta, juya-u-bicha _i sasaka-k
cloth-ins.pl red.cover hill-dir-towards red.go.pl-pfv
‘After the Chapayecasi danced, they took off the masks, _i/*j put them in
front of the church, _i/*j covered their heads with cloths, and _i/*j went
towards the hill.’
The constructional schema in Table 14.1, adapted from the construc-

tional schema for English ‘conjunction reduction’ (Van Valin 2005: 231),
captures the essential features of ‘topic-chain’ clauses in Yaqui and some
other languages. Note that the semantic representation of each non-
initial clause contains a lexically unfilled, obligatory co-indexed argu-
ment. Thus, there is a special linking requirement that involves highly
topical PSAs as pivots and obligatory coreference with the initial
pragmatic controller.
The next section examines how the linking algorithm works for adverbial
clauses. I show that some adverbial relations can be expressed by ad-
subordination, but others are coded by non-subordinated linkages. RRG
assumes that the semantic relations of the adverbial subordinate clause to
the unit it modifies are the same as those of a peripheral PP modifying the
core or clause. For example, since a PP like after the party expresses the
temporal setting of the event expressed in the core, it then follows that a
clause like after she arrived at the party also occupies the peripherycore . Reason
clauses, such as Chris was angry because of Pat’s insults, provide the reason,
cause or condition of the main clause as a whole; as a result, they occupy the

Table 14.1 Constructional schema for Yaqui ‘topic chains’
C ONSTRUCTION : Reduction constructions in Yaqui

S YNTAX :
Juncture: clausal
Nexus: cosubordination
Construction type: juxtaposition
[C L A U S E [ C O R E RPi [N U C . . .]. . .]. . .]1, [C L A U S E [ C O R E -i [N U C . . .]. . .]. . .]2,
[C L A U S E [ C O R E -i [N U C . . .]. . .]. . .]n
Unit template(s): step 4 in (8)
PSA: clause 1: variable syntactic controller ¼ pragmatic controller
clause 1 þ n: variable syntactic pivot ¼ pragmatic pivot
Linking: controlled argument in 1 þ n ¼ pragmatic pivot (steps 3a, c2)
M ORPHOLOGY :
CLM: none
S EMANTICS :
Sequence of events sharing a common primary topical participant
P RAGMATICS :
Illocutionary force: shared across all conjuncts
Focus structure: predicate focus in all conjuncts
peripheryclause . Hence, ad-subordination mostly follows the basic linking

algorithm in (8) and (9); the real challenge comes from non-subordinate core
junctures with their obligatory sharing of a core argument (Van Valin
2005: 225).
14.4 The Linking Algorithm in Adverbial (Adjoined)

Clauses
In the analysis that follows, adverbial relations are organized into five
groups: concessive and conditional clauses (14.4.1), reason clauses (14.4.2),
temporal clauses (14.4.3), purpose clauses (14.4.4), and manner and means
clauses (14.4.5). Adverbial relations may be established at different levels of
the clause structure; the relationships at the bottom of the semantic hier-
archy usually modify the clause, temporal and purpose clauses may modify
the core, and manner and means can usually modify the nucleus.
14.4.1 Concessive and Conditional Clauses

Concessive and conditional are the loosest semantic relations in terms of
the semantic hierarchies presented in (7). Concessive clauses express a
relationship of general incompatibility and counter-expectation between
two situations; in (12a), the content of the main clause holds unexpectedly,
given the content of the linked clause (Van Valin 2005: 207). In conditional
clauses, the protasis (the if-clause) describes a condition for the realization
of the situation expressed in the apodosis (the main clause); it can refer to a
present, past or future situation (12b), to an imagined situation that might

hypothetically happen, as in If he gets the job, we’ll all celebrate, or that did not
happen, as in If you had been at the party, you would have seen her new boyfriend
(Hetterle 2015: 49). In the English constructions in (12), each clause links
separately, but the complete construction encodes two semantically asso-
ciated states of affairs. Each association is formally captured in the (sim-
plified) logical structure of the construction by different connectors or
CLMs with lexical content: IN . SPITE . OF ′ for concessive (12a) and for
conditional (12b).
(12) English concessive and conditional clauses (Van Valin 2005: 207)
a. Bill made it to work, even though it was snowing heavily.
a′. [Bill made it to work] IN . SPITE . OF ′ [it was snowing heavily]
b. If it rains, we won’t be able to have a picnic.
b′. [If it rains] [we won’t be able to have a picnic]
These adverbial constructions do not entail operator dependency; they

can even have different IF values, as in if it rains, are we still having a picnic?
They do not demand argument sharing either. When the two PSAs are
coreferential, like in Were Fred to leave now, he would look like a fool, the
linked unit overtly expresses all its core arguments. The ordering distri-
bution of main and adverbial units may also vary (e.g. Even though it was
snowing heavily, Bill made it to work for (12a)). These key properties of the
constructions follow from there being ad-subordination at the clause
level. Because of their semantic association to the main unit, conditional
and concessive clauses occur in the peripheryclause . For the purposes of
linking, ad-clausal subordination is captured by step 5d1b in the seman-
tics-to-syntax algorithms in (8). Thus, the main clause and the linked
clause are linked independently of each other, just as if each were a
simple sentence on its own. When assigning arguments (and clauses) to
positions in the syntactic representation of the sentence, step 5d1b
assigns the relevant CLM þ clause to the peripheryclause . A simplified
syntactic representation of ad-clausal subordination in English is pre-
sented in Figure 14.2.
SENTENCE
PERIPHERY CLAUSE
CORE
CLM CLAUSE
CORE
‘If it rains, we won’t be able to have a picnic’ in (12b)

Figure 14.2 Conditional ad-clausal subordination in English

14.4.2 Reason Clauses

Reason clauses express a cause or reason that brings about the situation
expressed in the main clause. The logical structure for the reason clause in
(13a) includes the BECAUSE ′ connector.
(13) English reason clauses (Van Valin 2005: 207)

a. Kim berated Pat because she kissed Chris.
a′. [Kim berated Pat] BECAUSE ′ [she kissed Chris]
Some languages have several types of reason clauses. For instance, Cavineña
distinguishes cause clauses from reason clauses (Guillaume 2008). The ¼ra
clause in (14a) expresses the cause of the main event, whereas the ¼tibu clause
in (14b) gives the reason for the occurrence of the main event. Unlike cause
clauses, the linked verb in reason clauses can take the full range of verbal
morphology, and there are no coreference restrictions.
(14) Cavineña (Pano-Tacanan; Guillaume 2008: 719, 731)

a. E-tsaka uje-da [ju-ya aje¼ra]
npf-leg painful-asf be-impfv walk¼cause
‘My legs hurt from walking (so much).’
b. Aama! [mi-ra¼dya iye-wa¼tibu] duju-kwe
not.exist 2sg-erg¼foc kill-pfv¼reason take-imp.sg
‘No! Since you killed it (a caiman), you take it (not us)!’
In the interclausal semantic relations proposed by Van Valin (2005:

206–207), reason shows a closer semantic association when compared to
conditional and concessive because of its causal meaning (i.e. the causal
hierarchy in (7b)). However, languages may vary the syntactic linkages used
to express these relationships. For instance, in Cristofaro’s study (2003:
229), reason outranks condition in the lack of tense, aspect and mood
distinction, as well as the use of case marking/adpositions on the linked
verb, but the two are ranked together in terms of the coding of partici-
pants. Hetterle (2015: 143) comments that in Chinese, preposed yı̄nwèi
cause clauses and rúguŏ conditional clauses are under the scope of main
clause interrogation, while suı̄rán concession clauses and jı̀rán reason
clauses are not; this suggests that cause and conditional clauses show a
tighter syntactic linkage when compared to concessive and reason
clauses.6 Diessel and Hetterle (2011) also suggest that reason clauses in
English, German and Japanese function as independent assertions that are
only loosely combined with the associated main clause. Reason clauses, but
not conditional clauses, are often intonationally separated from the
semantically related clause (even in their canonical position). Therefore,
the degree of syntactic integration of the linked unit that expresses con-
cessive, condition and reason may vary across languages. This is the case in
Yaqui. In Yaqui, reason and concessive clauses are structurally different
from conditional clauses. First, concessive and reason clauses are the only
clause types that make use of initial CLMs with lexical meaning: ella’apo

‘although, even though’ (15a), and kiali’ikun ‘because, so that’ (15b). Second,
the description of the event in the adverbial unit is fully independent in
terms of TAM, negation and IF. Third, concessive and reason clauses can
attract restricted discourse markers that occur on the left edge in inde-
pendent clauses and can have their own pre- or post-detached phrases
(PrDP, PoDP). Finally, the dependent PSA takes nominative case; nomina-
tive PSAs are completely banned in syntactically subordinate clauses in
Yaqui (Guerrero 2006, 2017).
(15) Yaqui concessive and reason clauses
a. Bempo Guayma-meu saja-ne [ella’apo empo im tawa-ne]
1pl.nom Guaymas-dir.pl go.pl-pot clm 2sg.nom here stay-pot
‘We will go to Guaymas although you stay here.’
b. Inepo into in jaboii¼tua, tei tu’isi emo jala’eka-n,
1sg.nom and 1sg.gen grandpa¼int 1pl.nom good self friend.be-ipfv
[kiali’ikun ne au¼waate into ne a-et ai ¼ baisae-Ø
clm 1sg.nom 3sg.obl¼miss-prs and 1sg.nom 3sg-loc 3sg.acc¼thank-prs
si’ime-tai ]
everything-acc
‘As for me and my grandpa, we were very good friends, so I miss him and I thank him for
it, for everything.’ (Buitimea; regreso: 37)
As for conditional clauses, three syntactic structures have been identified

so far, all of which take the underspecified CLM -o. In addition to -o, one
structure takes the Spanish loanword si ‘if’ (16a), and another takes the
special verb marker -tek ‘conditional’ (16b); the last example (16c) does not
take any extra marking. The linked verb is usually unmarked for TAM values
or is marked by the perfective suffix -k only, hence there are some operator
restrictions. Additionally, the dependent PSA empo ‘you’ in (16a–b) takes the
form of a nominative pronoun, but in (16c) it occurs in its accusative form,
enchi ‘you’.
(16) Yaqui conditional clauses

a. [Si empoi mam su’utoja-o]
clm 2sg.nom arm.pl.acc release-clm
empoi che’a juni nee a¼kotta-tua-ne
2sg.nom more even 1sg.acc 3sg.acc¼break-cause-pot (Guerrero; Lalo: 57)
‘If you release the arms, you will make me break it [the elbow] even more.’
b. [Empo teopo-u lionok-bae-tek-o], empo emo jin-tua-ne
2sg.nom church-dir pray-want-cond-clm 2sg.nom self cover-cause-pot
‘If you want to pray in the church, you have to cover yourself.’
c. Inepo enchi ania-ne [enchi junuen ’ea-o]
1sg.nom 2sg.acc help-pot 2sg.acc thus think-clm
‘I will help you if you want (me to).’ (Dedrick and Casad 1999: 393)
Therefore, reason and concessive clauses in Yaqui link together two independ-
ent units in terms of an initial CLM, operator and argument coding; in (15b),
there are two complete sentences, each with its own detached position: there
is a topical element in the PrDP of the first component, as for me and my

TEXT
SENTENCE CLM SENTENCE
PrDP CLAUSE CLAUSE CLM CLAUSE
CORE CORE CORE

RP
Inepo into in jaboii=tua, tei tu’isi emo jala’eka-n, kiali’ikun ne au=waate into ne a-et ai=baisae-Ø si’ime-tai
‘As for me and my grandpa, we were very good friends, so I miss him and I thank him for it,
for everything’ in (15b)
Figure 14.3 Simplified structure of reason relations as sentential coordination in Yaqui
grandpa, and there is an anti-topic element in the PoDP of the second unit,
everything (sentential coordination). In contrast, conditional clauses show two
key features signalling structural dependency: the presence of a final, bound
and underspecified CLM, and operator dependency; (16c) also satisfies the
need for an accusative dependent PSA (ad-clausal subordination).
A simplified syntactic representation of (15b) is offered in Figure 14.3.
In order to capture the semantic association of ad-core (14a) and ad-
clausal (12, 13, 14b, 16) adjuncts, Van Valin (2005: 229) revised step 4 in
the syntax-to-semantic algorithm in (9). The relevant modifications are in
boldface in (17). Step 4b allows us to represent concessive, conditional and
reason PPs or clauses in the periphery. Sentential coordination for the
Yaqui constructions in (15) does not require any modification to the
linking algorithm.
(17) Revision of step 4 in the syntax-to-semantics linking algorithm (Van Valin 2005:
229)
4. If there is an adjunct phrase in a periphery,
a. and if it is in the peripherycore , then retrieve the logical structure of the
predicative adposition from the lexicon, insert the logical structure of
the core as the second argument in the logical structure and the object
of the adposition as the first argument,
b. and if it is in the peripheryC L A U S E , then link the adjunct PP or clause
logical structure to the matrix logical structure via the semantic
representation of the predicative adposition or CLM.
14.4.3 Temporal Relations

Next on the semantic continuum are temporal relations. Temporal expres-
sions provide the time specification for the state of affairs expressed in a
sentence. They do so indirectly and in relation to another moment, situation
or event (Declerck 1997). Typological studies have shown that temporal
clauses tend to show a higher degree of integration with the main clause

than conditional and reason clauses. For instance, temporal relations (but
not reason relations) involve the predetermination of time and aspect values
of the linked state of affairs, and might often include reduced verb forms
(Cristofaro 2003: 173). Across languages, temporal relations can be expressed
by different syntactic means (adverbs, adpositional phrases, finite and non-
finite clauses), and can appear to the left or to the right of the core they
modify (i.e. as modifiers in the periphery of the core).
The semantic association between the two units can be captured by CLMs
with lexical content. For instance, ‘before’ expresses posterior relations with
respect to the main event, ‘after’ introduces anterior relations, and ‘while’
introduces co-temporal events. In Spanish, the prepositions antes (de) ‘before’
and después (de) ‘after’ can take phrases, cores or clauses as objects (Guerrero
et al. 2017). In (18a–c), there is a subjectless, non-finite infinitival unit. As
mentioned before, infinitival verbs are a type of non-subordinate core that
depends on the main verb in terms of TAM operator. However, the infinitival
unit in temporal clauses can be independently negated from the main
clause, as in Sam se durmió después de no cenar ‘Sam fell asleep after not having
dinner’. Commonly, the PSA of the main core (Sam) controls the referential
identity of the missing argument (pivot) within the linked core, but in (18c)
each core has its own PSA. This means that argument sharing is optional.
Hence, since the two cores can be independently negated, and they can but
do not necessarily share an argument, (18a–c) involve core coordination.
There is an additional sequential construction. In (18d), the linked unit is
introduced by the complex CLM después de que ‘after that’ and the two units
are independent of each other in terms of TAM operators at the level of the
clause and argument coding. Therefore, (18d) takes a modifying clause at
the periphery of the core, hence ad-core subordination. In Spanish, the
unmarked position for temporal adverbs, phrases and clauses is postverbal,
though they may also appear at the beginning, as in antes de acostarse, Sam
besó a Pat for (18a). The simplified logical structure for sequential construc-
tions is the same as that of a peripheral PP modifying a core and is valid for
the two juncture–nexus.
(18) Spanish temporal clauses

a. Sami besó a Patj [_i antes de acostarse]
‘Sam kissed Pat before lying down.’
a′. be-before′ ([INGR lie down′ (xi)], [do′ Sami, [kiss′ (Sami, Pat)]])
b. Sami se durmió [después de__i besar a Pat]
‘Sam fell asleep after kissing Pat.’
b′. be-after′ ([do′ (xi, [kiss′ (xi, Pat)])], [do′ (Sami, [sleep′ (Sami)]])
c. Pat vio la película [después de dormirse Sam]
‘Pat saw the movie after Sam fell asleep.’
c′. be-after′ ([do′ (Sam, [sleep′ (Sam)])], [see′ (Pat, movie)])
d . Pat vio la película [después de que Sam se durmió]
‘Pat saw the movie after Sam fell asleep.’
d′. be-after′ ([do′ (Sam, [sleep′ (Sam)])], [see′ (Pat, movie)])

SENTENCE
CLAUSE
CORE PERIPHERY
PP
COREp
NUCp CLAUSE
Pat vio la película P
después de que Sam se durmió

‘Pat saw the movie after Sam fell asleep’ in (18d)
Figure 14.4 Sequential temporal relations as ad-core subordination in Spanish
Peripheral subordination at the core level does not require any revision
of the linking algorithm, as it is captured by the steps in (5d1a) in the
semantics-to-syntax linking (8): the predicative preposition takes either a
core or clausal unit and, as a whole, is attached to the periphery of the core.
When linking syntax to semantics, step 4a yields the representation for
spatial and temporal circumstances, based on the predicative adposition.
A simplified syntactic representation for the sequential construction in (18d)
is presented in Figure 14.4. This is another example of asymmetrical linkage
since a linked clause is contained within a sub-clausal unit.
Core coordination junctures do require revision of the linking algorithm,
especially when the main actor controls the pivot in the linked unit, as in
(18a–b). In order to capture the fact that there is a syntactic argument slot
missing in the linked unit, the template selection principles in the seman-
tics-to-syntax linking (8) were revised, and a new principle (step 4b) was
added for non-subordinate core junctures. Step (4b1) specifies that the
linked unit is a semantic argument of the matrix verb, as in Chris tried to
see Pat. However, temporal clauses do not fulfil a semantic slot in the logical
structure of the main clause. The nature of Spanish temporal infinitival
cores in (18a–b) can be captured by step (4b2).
(19) Modifications to the semantics-to-syntax linking (Van Valin 2005: 262–263)

4. Syntactic template selection principle
b. Universal qualification of the principle in (a): The number of slots in a
core is reduced by 1 if:
1. it is the matrix core in a non-subordinate core juncture in which the
linked core is a semantic argument of the matrix verb, and/or7
2. it is the linked core in a non-subordinate core juncture.
The statement in step 4b does not specify which syntactic slot is missing,
since that is a construction-specific feature. The syntax-to-semantic linking
needs some modifications too. The relevant adjustments apply after the

logical structure of the clause has been retrieved (step 2), and the linking of
macrorole and non-macrorole core argument has taken place (step 3). Again,
step 4b can capture the fact that there is a syntactic argument missing from
the linked core which must be interpreted as being the same as one of the
syntactic arguments from other positions. The logical structures for (18a)
and (18b) contain a lexically unfilled co-indexed argument x, which is
controlled by the PSA of the main clause (Sam).
(20) Modifications to the syntax-to-semantic linking (Van Valin 2005: 280–281):

4. In non-subordinate core junctures, one of the arguments of the matrix
core must be linked to an argument position in the embedded logical
structure:
a. if the matrix predicate is a control verb, this follows the theory of
obligatory control: causative and jussive verbs have undergoer
control, while all other (M-)transitive verbs have actor control;
otherwise,
b. if the matrix predicate is not a control verb, then link the unlinked
syntactic argument in the matrix core to the logical structure
argument position of the pivot of the linked core.
That is, all temporal clauses in (18) semantically modify the core of the main
unit, but only the linked unit in (18d) occupies a peripheral position; the
non-finite linked units in (18a–c) yield a non-subordinate core juncture.
A simplified representation for core coordination expressing sequential
relations in Spanish is presented in Figure 14.5.
Depending on their role in discourse, temporal clauses can also occupy
the pre- or post-detached position, with or without a prominent pause
between the two units (Guerrero et al. 2017). In a sentence like Ricardo
S fue expulsado en 1906, después de pasar dos años removiendo el caso de la muerte
de DG ‘Ricardo S was expelled in 1906, after spending two years investigating
the death of DG’, there is a non-finite core, and the missing argument is co-
indexed with the main actor. In this construction, there are two temporal
expressions, the PP ‘in 1906’, and the ad-clause introduced by the prepos-
ition después de ‘after’. Because there is another core modifier, I suggest that
the after-clause does not modify the core, but is linked to the sentence
directly in the post-detached position (PoDP). Notice that there is another
SENTENCE
CLAUSE
CORE CLM CORE
Pat vio la película después de besar a Pat

‘Sam fell asleep after kissing Pat’ in (18b)
Figure 14.5 Sequential temporal relations as core coordination in Spanish

SENTENCE
CLAUSE PoDP
CORE PERIPHERY PP
PP CORE
COREp NUCp CORE
Ricardo S fue expulsado NUCp RP PRED
PRED p
p
en 1906, después de
pasar dos años removiendo el caso de la muerte de DG
‘Ricardo S was expelled in 1906, after spending two years investigating the death of DG’
Figure 14.6 Simplified structure of sentential subordination in Spanish
temporal PP modifying the core. The representation of this construction is

presented in Figure 14.6. According to the linking algorithm in (8), the
temporal PP is captured by the steps in 5d1a, while the temporal clause is
placed at the PoDP following the steps in 5d3.
Not all temporal clauses are introduced by predicative prepositions. Gen-
eral when-clauses are particularly interesting because the CLM does not
specify the temporal sequence between the main and the adverbial units.
For English, it has been said that when-clauses can express any temporal
relation, although they strongly encode temporal overlap ‘events even
though the exact extent of the overlapping is subject to variation’ (Cristofaro
2003: 159; Diessel 2008: 474). The same has been said for Spanish construc-
tions introduced by cuando ‘when’ as in (21). In corpus analysis, cuando-
clauses turn out to be the most productive temporal clause type (Gerardo-
Tavira 2018). The examples in (21a–b) are from Conti (2012), and (21c) is from
Gerardo-Tavira (2018).
(21) a. _i llámame [cuando _i llegues]

‘Call me when you arrive.’
b. [Cuando _i llegó a Tepic], _i quedó fascinado con el verdor del paisaje
‘When he arrived at Tepic, he was fascinated by the green landscape.’
c. [Cuando _iþj éramos novios], _i mej llegaba a visitar de leva traslapada
‘When we were dating, he would visit me wearing his finest clothes.’
There is an assumed correlation between the use of a comma (as a correl-

ate of an intonational break) and the extra-clausal position of the temporal
adjunct (Van Valin 2005: 192–193). Conti (2012) recognizes this correlation
and proposes that the clausal juncture in (21a) occupies the periphery of the
clause, while that in (21b–c), with a comma after the linked unit, is outside
the clause, and is linked directly to the sentence in the PrDP. Since cuando/
when are not predicative prepositions, but CLMs with lexical content, one

may wonder what the logical structure of the constructions in (21) looks
like. The Interclausal Semantic Relation Hierarchy (Van Valin 2005: 206–207)
recognizes temporal circumstances as the spatial or temporal parameters of
an event (e.g. Kim saw Pat after she arrived at the party) and temporally ordered
states of affairs. There are simultaneous states of affairs, where one state of
affairs is temporally coterminous with another, like in (22a), as well as
sequential states of affairs, in which one state of affairs follows another
temporally, with or without any temporal overlap, as in (22b). Note that the
linkage markers do not have lexical meaning but resemble some sort of
conjunction/connector. That is, the semantic association between the units
cannot be captured by the CLMs, but instead depends on the pragmatic
context.
(22) a. Max was dancing, and at the same time Susan played the piano.
a′. [do′ (Max, [dance′ (Max)])] ^ [do′ (Susan, [play′ (Susan, piano)])]
b. Juan finished reading the newspaper, and then Carlos walked into the room.
b′. [do′ (Juan, [read′ (Juan, paper)])] & [do′ (Carlos, [walk′ (Carlos) & INGR be-at′
(room, Carlos)])]
14.4.4 Purpose Relations

Next in the semantic continuum are purpose relations, in which one action
is done with the intent of realizing another state of affairs. Unlike reason,
conditional and temporal relations, purposive situations show a strong
preference for argument sharing (recall the Kokama examples in (6)). First
Cutrer (1993), and then Van Valin (2009), proposed two types of purposive
constructions in English based on the lexical manifestation of the controllee
in post-nuclear position: ‘pure’ purpose clauses that exhibit an obligatory
missing syntactic argument in the linked unit (23), and rationale clauses, in
which there is not a missing argument, but a co-indexed pronoun (24). Note
that the controllee must be associated with the immediately post-nuclear
argument (undergoer) in the infinitival core. The pre-nuclear argument
(actor or PSA) may be the same (23a–b, 24a–b) or different (23c, 24c). The
data comes from Cutrer (1993).
(23) a. Johni caught a fishj [_i to eat _j for dinner]

b. Johni gave a tapej to Maryk [_k to listen to_j]
c. Johni built a chestj for his sisterk [_k to put her clothes in_j]
(24) a. Johni bought Maryj a bookk (in order) [ _i to please herj]

b. Johni sang the childrenj a lullaby (in order) [ _i to calm themj]
c. John bought the turkeyi (in order) for his wifej [ _j to cook iti]
In Cutrer’s analysis (1993: 177), purpose clauses are a type of core juncture,
and therefore, there must be an argument shared between the two cores. It
is possible to have two control relations, as in (23b), but only one is obliga-
tory: the postverbal gap must be obligatorily controlled by the main theme.

In opposition to this, rationale clauses are considered a type of clause

juncture, and hence there are no obligatorily shared arguments between
the two clause units; instead, there is a post-nuclear pronoun when transi-
tive, as in (24). Unlike the linked undergoer, the actor gap has a non-
obligatory control relation, since it may be filled or unfilled. In (23a–b) and
(24a–b), the main actor controls the identity of the missing agent in the
linked unit; in (23c) and (24c) the controller is a non-macrorole oblique core
argument. Thus, despite the semantic similarities of the two constructions,
their syntactic properties are different.
However, the position and the lexical coding of the controllee are subject to
language-specific conditions. In order to avoid structural assumptions in
particular languages, it is highly recommended to describe the function of
the controller and controllee in terms of actor/agent (e.g. pre-nuclear in
English), and undergoer/theme, recipient (e.g. post-nuclear in English), and
the lexical coding of the controllee in terms of structural and inherent
control (Guerrero 2012, 2013). For instance, there are three purposive con-
structions in Yaqui (Guerrero 2011, 2013, 2017): the first type (25a) lacks CLM;
the second type (25b) is marked by the general CLM -kai, and the third type
(25c–d) is marked by the postposition -betchi’ibo ‘for/in order to’. The linked
core in (25a–b) fully depends on the main core for operator values and cannot
be independently negated. In contrast, the linked core in (25b–c) can be
unmarked or marked by the potential suffix -po, and it can be negated.
Furthermore, all purposive constructions demand argument sharing, but
obligatory shared arguments make use of a special lexical coding. Thus, there
is structural control (gap/missing syntactic argument) when the controller is
the main actor (25a–c), and inherent control (co-indexed pronoun) when the
controller is the main undergoer (25d). Since Yaqui is a head-final language,
both the controller and the controllee must be pre-nuclear.8
(25) Yaqui purpose clauses
a. Peo-Øi a’abo _i ji’i-bwa-se-k
Peter-nom here something-eat-move.purp.sg-pfv
‘Peter came here to eat.’
a′. [do′ (Peter, [MOTION′ (Peter)])] ^ [do′ (Peter, [eat′ (Peter, something)])]
b. [Kafe-ta _i ji’i-pea-ka] nei ino piaroa-k
coffee-acc drink-want-clm 1sg.nom 1.self borrow.money-pfv
‘I borrowed some money in order to drink coffee.’
b′. want′ (I, [do′ (I, [drink′ (I, coffee)])] ^ DO [do′ (I, [borrow′ (I, money)])]
e CAUSE [do′ (I, [drink′ (I, coffee)])]
c. U o’ou-Øi juya-u siika [mas-ta _i me’e-betchi’ibo]
det man-nom hills-dir go.sg.pfv deer-acc kill.sg-clm
‘The man went to the hills in order to kill the deer.’
c′. want′ (man, [do′ (man, Ø) CAUSE [BECOME dead′ (deer)]) ^
DO [do′ (man, [move′ (man)]) & INGR be-at′ (hills, man)]]
d. U maso-Øi bwite-k [u-ka o’ou-taj kaa ai_me’e-ne-betchi’ibo]
det deer-nom run.sg-pfv det-acc man-acc neg 3sg.acc¼kill.sg-pot-clm
‘The deer ran quickly in order for the man not to kill it.’

d′. [do′ (deer, [run′ (deer)] PURP [do′ (man, Ø) CAUSE [NOT BECOME dead′ (deer)]]]
e. Min-Øi yoi-taj kaba’i-mkreuwa-k [_j amk¼wiria-ne-betchi’ibo]
Fermín-nom foreigner-acc horse-pl lend-pfv 3pl.acc¼feed-pot-clm
‘Fermín lent the foreigner the horses in order for him to feed them.’
e′. [do′ (Fermín, [lent′ (Fermín, horses)] CAUSE [BECOME have′ (foreigner, horses)]] PURP
[do′ (foreigner, [feed′ (foreigner, horses)])]
Notice that Yaqui does not distinguish between purpose (25a–d) and
rationale clauses (25e) based on an obligatory syntactic gap, since a referen-
tial controlled undergoer must be overt. In the last example, the two cores
share both the theme and the recipient, which are lexically expressed as
core arguments of the main unit; in the linked unit, the recipient yoi
‘foreigner’ controls the actor PSA of the linked verb ‘feed’, hence there is
structural control (missing syntactic argument), while kaba’im ‘horses’ con-
trols the undergoer of ‘feed’, and so there is inherent control (co-index
pronoun). Since there are unfilled/filled syntactic arguments in the linked
core which need to be obligatorily controlled by a core argument in the
main unit, Yaqui purpose clauses behave as a kind of control construction
(Guerrero 2017).9 As a matter of fact, the special lexical coding of the
controllee is exactly the same in purpose clauses and constructions taking
actor and undergoer control verbs. In Yaqui, actor control verbs such as
modal, phasal and desiderative constructions demand a cover controllee in
the linked unit (26a), whereas undergoer control verbs like causative and
jussive constructions demand an overt controllee, as depicted in (26b). In
purpose, causative, and jussive constructions, a missing argument co-
indexed with the main theme would be ungrammatical.
(26) Yaqui inherent actor and undergoer control verbs

a. Lupe-Øi kaa _i yi’i-pea-Ø.
Lupe-nom neg dance-int-prs
‘Lupe does not want to dance.’
b. Empoi Goyo-taj sawe-k [tekil-ta aj¼tekipanoa-ne-’u]
2sg.nom Goyo-acc order-pfv land-acc 3sg.acc¼work-pot-clm
‘You ordered Goyo to work the land.’
Comanche, another Uto-Aztecan language, also shows noteworthy

controller–controllee patterns. According to McDaniels (2014), in rationale
clauses either an object gap (27a) or a subject gap may occur (27b), or neither
(27c), but not both. Thus, the lexical manifestation of the controllee varies in
rationale clauses (it can be overt or covert), and this variation does not
depend on its function inside the linked unit. Purpose clauses show the
same controller–controllee patterns, but the linked unit is marked by the
accusative suffix -a (i.e. a nominalized purpose clause). In (27d) there is a
missing syntactic argument serving as the theme of the dependent verb,
which is obligatorily controlled by the main theme. According to McDaniels
(2014: 79), Comanche purpose and rationale constructions consist of subor-
dinated clauses at the core level.

(27) Comanche (Northern Uto-Aztecan; McDaniels 2014: 75, 73, 77, 83)
a. Yuhu-noko-pī-aj nīi tīmī-i [nīi _j tīhka-tui]
fat-bake-res-acc 1p(s.clitic) trade-pfv 1p eat-unr
‘I bought frybreadj in order for me to eat _ j’.
b. S-u-tī-sei peekwij kwīhī-n [uj _i tīhka-tui]
def-3p-nom-dm fish catch-pfv 3p(o.clitic) eat-unr
‘Hei caught a fish (in order) _ i to eat it’.
c. S-u-tī-sei u-hkupa uj tīīki [u-hkaj pīi napinai-tui]
def-3p-nom-dm 3p-inside 3p(o. clitic) put 3p-acc 3p.co save/select/stash-unr
‘Hei put it in there (in order) for himi to set it aside’.
d. S-u-tī-sei tī-poo-pī-aj [pīi _j niha-tui-a] yaa-tī
def-3p-nom-dm unsp.o-make.mark-res-acc 3p.co.gen read-unr-acc take-ipfv
‘He is taking a booki to read _i’.
The co-indexed pronouns in purpose clauses marked by -betchi’ibo in Yaqui

and the rationale clauses in Comanche do not require any modification to the
linking algorithm, but the obligatory control relation observed in purpose
clauses in (25a–c) mainly follows the modifications introduced in the seman-
tics-to-syntax algorithms in (19) for non-subordinate core junctures, but some
revisions are needed for the syntax-to-semantics linking (20). Typological
studies have found that the argument structure in the linked unit is not
logically predetermined by the meaning of purpose, ‘although . . . typical
purposive situations are bound to their matrix event via agent-binding’
(Schmidtke-Bode 2009: 50). Schmidtke-Bode’s work on purpose clauses reveals
significant trends for coreference in general, and same-subjects in particular.
The second most common pattern involves the main theme. In Yaqui purpos-
ive constructions, coreference is not an option but a must. Therefore, in order
to capture this complexity, additional information may be needed with
respect to the semantics of the clause, and the fact that the lexical manifest-
ation of the controllee is subject to language-specific restrictions. As is, the
step in 4 in (20) establishes that there is a controller–controllee relation
among an argument in each core inside a non-subordinate core juncture;
step 4a is relevant for those constructions including a control verb as a matrix
verb; step 4b tries to capture constructions without coreferential control.
However, Yaqui purpose clauses do not have a control verb, and yet the
selection of the controller follows the theory of obligatory control.
I therefore suggest adding the specification in the new step 4c, including
the referential control hierarchy introduced in (7e).
(28) Revision of the syntax-to-semantics linking algorithm (Guerrero 2013: 17)

4. In non-subordinate core junctures, one of the arguments of the matrix
core must be linked to an argument position in the embedded logical
structure:
a. If the matrix predicate is a control verb, then follow the theory of
obligatory control constructions: causative and jussive verbs have
undergoer control, while all other (M-)transitive verbs have actor
control, or

b. If the matrix predicate is not a control verb, then link the unlinked
syntactic argument in the matrix core to the logical structure
argument position of the pivot of the linked core, otherwise
c. If the non-subordinate core juncture expresses a purpose relation, and
1. if the construction demands structural control, then the linked
argument must be unfilled (e.g. syntactic controllee in English,
Lakhota, Dyirbal); otherwise,
2. if the construction demands inherent control, then the linked
argument can be filled by a bound pronoun, clitic or agreement-
inflection, (e.g. semantic controllee in Acehnese, Chuj, Yaqui,
Toqabaquita)
3. the identity of the controllee may follow the referential control
hierarchy:
the referential identity of an obligatory controlled argument is
determined by the main actor > the main undergoer > another
main core argument > there is no control relation.
Instead of ‘pure’ syntactic control constructions, the steps in 4c try to

capture the referential controller–controllee patterns found in purpose and
rationale clauses across languages (see Guerrero 2012, 2013). The controllee
can be covert and show structural control, as depicted by the actor gap in
(23), (24), (25a–c), (26a), and (27b), and the undergoer gap in (23), (27a, d); or it
can be overt and yield inherent control, as in the co-indexed pronoun for the
undergoer in (24), (25d–e), (26b), (27c–d). Step 4c3 seeks to mirror the most
common coreferential patterns observed cross-linguistically, as well as
Cutrer’s (1993) description.
In sum, purposive constructions may be expressed by different syntactic
linkages in particular languages. Simplified representations for core
cosubordination (25b) and core coordination (25c) in Yaqui are presented
in Figure 14.7 and Figure 14.8, respectively.
The constructional schema in Table 14.2 tries to capture the syntactic and
semantic properties of the Comanche rationale clause in (27a). Here, there
are two referential control relations. There is inherent control (co-indexed
pronoun) involving the main PSA and the linked actor; and there is struc-
tural control (gap) involving the main and the linked theme.
SENTENCE
CLAUSE
CORE
CORE CLM CORE
[Kafe-ta _i ji’i-pea-ka] nei ino piaroa-k

‘I borrowed some money in order to drink coffee’ in (25b)
Figure 14.7 Simplified structure of purposive core cosubordination in Yaqui

SENTENCE
CLAUSE
CORE CORE CLM
U o’ou-Øi juya-u siika [mas-ta _i me’e-betchi’ibo]

‘The man went to the hills in order to kill the deer’ in (25c)
Figure 14.8 Simplified structure of purposive core coordination in Yaqui
Table 14.2 Constructional schema for ‘I bought frybread in order for me to eat’ in
(27a)
C ONSTRUCTION : ‘Rationale’ construction in Comanche

S YNTAX :
Juncture: core
Nexus: coordination
Construction type: adverbial non-subordination
[C L A U S E [ RP [C O R E 1 . . .]] PURP [ RP [C O R E 2 . . .] (RP)]]]
Unit template(s): default
PSA: the controller is the matrix actor in CORE 1
the controllee is the linked actor in CORE 2
Linking: the number of core slots is maintained, i.e. inherent control (4c2)
Non-PSA: the controller is the matrix undergoer in CORE 1
the controllee is the linked undergoer in CORE 2
Linking: the number of core slots is reduced by 1, i.e. structural control (4c1)
M ORPHOLOGY :
Verb: non-finite
PSA Controllee: filled
Non-PSA Controllee: unfilled
CLM: none
S EMANTICS :
One action is done with the intent of realizing another state of affairs
The identity of the controllee follows the referential control hierarchy (4c3)
P RAGMATICS :
Illocutionary force: unspecified
Focus structure: unspecified
14.4.5 Manner and Means Relations

Finally, the semantic relations at the left end of the semantic scale denote
modifying subevents such as manner, means, motion and position. Most
studies on adverbial subordination do not mention manner/means (Ford
1993; Cristofaro 2003; Givón 2001), while some only consider manner/simi-
larity relations (Dixon and Aikhenvald 2009). Others include both and high-
light the verb form and an overt CLM (Hengeveld 1998; Van Valin 2005); and
some others group manner, time and location together (Thompson et al.
2007). In English, manner and means relations are usually coded by the -ing
verb form; an example is given in (29a). In the Spanish example in (29b),
taken from Conti (2012), there is a gerund verb form ending in -ando.

(29) a. John cut the bread by slicing it

b. El pájaro volaba imitando el aleteo de su madre
‘The bird flew imitating the wingbeats of its mother.’
In Yaqui, these adverbial relations are coded by the general CLM -kai. In the
constructions in (30), -kai introduces the manner in which a motion event is
carried out (30a), and the means by which an action is carried out (30b). The
example in (30c) expresses some sort of positive/negative circumstance
around the main action. The entire clause describes a complex event in
which the main actor is engaged. Because -kai is a general CLM, the actual
semantic relation between the two units is recovered from the discourse
context, the nature of the events, and world knowledge (e.g. two ongoing
events involving the actor; a functional answer for why). In the logical
representations, ^ represents two simultaneous states of affairs. Note that
there is an unfilled position in the second component co-indexed with the
main PSA.
(30) Yaqui manner and means clauses

a. [junama’a _i si tebae-koko-ka] te ama jooka
over.there int hungry-die.pl-clm 1pl.nom there live-prs
‘We live over there, starving to death.’ (Silva; HVH: 213)
a′. [live′ (1pli, there)] ^ [starved′ (xi, there)]
b. Into¼nei ousi tomi-yo’o-k [ _i yeewe-kai]
dm¼1sg.nom a.lot money-win-pfv play-clm
‘And then, I won a lot of money by playing.’ (Buitimea; chapayeca: 83)
b′. [do′ (1sgi, [won′ (1sgi, money)])] ^ [do′ (1sg, [play′ (xi)])]
c. [Kat]¼e’ei [_i ji’i-bwa-ka] to’o-ne
neg.imp¼2sg.nom something-eat-clm lie-pot
‘You will go to bed without eating.’ (Silva; zorrillo: 13)
c′. [do′ (2sg, [sleep′ (2sg)])] ^ [NEG do′ (xi, [eat′ (xi, Ø)])]
Yaqui -kai clauses can be considered equivalent to an imperfective partici-

pial -ing clause (Givón 2001). However, -kai clauses are more restrictive than
English -ing forms in terms of control. While in the English construction in
(31a) the antecedent of the missing PSA of ‘coming’ can be ‘she’ or ‘he’, the
Yaqui clause in (31b) rules out the possibility of the main object as the
controller. The possibility for ing-infinitival clauses with accusative/genitive
subjects (31c) is also disallowed.
(31) a. She saw him coming out of the library

b. Inepoi Joan-taj bicha-k [teopo-po _i yeu¼sim-kai]
1sg.nom John-acc see-pfv church-loc out¼go.sg-clm
‘I saw John leaving the library.’
c. Na’a tajkai-reo-Øi¼bea [_/*ai tajkai-m nenenka-ka] aui ania-n
dem tortilla-nmlz¼dm 3sg tortilla-pl red.sell-clm self help-ipfv
‘As for this tortilla maker, she helped herself by selling tortillas.’
The linked unit marked by -kai is not syntactically subordinated to the main
event, but adds a second predication to it. Several major features define

manner and means kai-clauses in Yaqui. The linked event can denote a state as
well as an activity predicative (manner is not limited to motion events, for
instance), and the two events are co-temporal (simultaneous). Additionally,
manner and means clauses are usually reduced units (intransitive verbs or
transitive verbs without referential core arguments), the verb must be
unmarked for TAM affixes and negation, and must generally occur at the
beginning or the end of the sentence without a pause. More importantly, they
are subjectless clauses. The main PSA (the actor) must control the identity of
the missing syntactic argument in the linked unit. In sum, this clause-linkage
type yields nuclear or core cosubordination depending on the valence of the
linked verb. In a nuclear juncture, the two nuclei combine to form a complex
nucleus with a single set of core arguments (30a–b). Figure 14.9 gives the
syntactic representation of nuclear cosubordination in (30b).10
In a core juncture, the two cores share one but not all arguments, as in
(30c) and (31b–c). Figure 14.10 gives the syntactic representation of core
cosubordination in (31c). The actual linking of this construction is the same
as the one in rationale/purpose clauses, in which there is a missing syntactic
argument in the linked unit co-indexed with the main PSA, and the linked
verb is unmarked for TAM suffixes.
SENTENCE
CLAUSE
CORE
RP NUCLEUS
NUC NUC CLM
Into=nei ousi tomi-yo’o-k [ _i yeewe-kai]

‘And then, I won a lot of money by playing’ in (30b)
Figure 14.9 Nuclear cosubordination in Yaqui
SENTENCE
CLAUSE
CORE
PrCs
CORE CLM CORE
RP
Na’a tajkai-reo-Øi=bea [_i tajkai-m nenenka-ka] aui ania-n

‘As for this tortilla maker, she helped herself by selling tortillas’ in (31c)
Figure 14.10 Core cosubordination in Yaqui

14.5 Conclusion
The main goal of this chapter was to examine how the linking algorithm
works in adverbial (adjoined) clauses. I have argued that the structural
linkages which express adverbial relations across languages are heteroge-
neous: some relations make use of looser syntactic combinations, while
others are encoded by tighter syntactic linkages. Furthermore, the same
semantic relation can show more than one syntactic linkage with different
degrees of semantic and syntactic integration. The theory of clause union in
RRG, based on juncture–nexus relations, conveniently accounts for these
structural complexities. On one hand, adverbial relations can be coded by
ad-subordination, as well as those relations expressed by coordination and
cosubordination. On the other hand, only ad-subordinated linkages occupy a
peripheral position at the level of the core or the clause, whereas non-
subordinated linkages demand a different syntactic representation.
References
Bickel, Balthasar. 1993. Belhare subordination and the theory of topic. In K.

Ebert (ed.), Studies in Clause Linkage (Papers from the First Köln-Zürich Work-
shop). Arbeiten des Seminars für Allgemeine Sprachwissenschaft (no. 12),
23–55. Zurich: University of Zürich.
Bickel, Balthasar. 2003. Clause Linkage Typology. Lecture series delivered at the
2003 International Role and Reference Grammar Conference, UNESP,
Brazil.
age: A multivariate analysis. In Bril (ed.), 51–104.
Bril, Isabelle. 2010. Clause Linking and Clause Hierarchy: Syntax and Pragmatics.
Conti, Carmen. 2012. Subordinación periférica y subordinación dependiente:
clasificación estructural de la subordinación adverbial en español. In R.
Mairal Usón, L. Guerrero and C. González-Vergara, El funcionalismo en la
teoría lingüística: La Gramática del Papel y la Referencia, 269–286. Spain: AKAL.
Conti, Carmen. 2021. Cosubordinación en español. Bern: Peter Lang.
Cristofaro, S. 2003. Subordination. New York: Oxford University Press.
Cutrer, Michelle. 1993. Semantic and syntactic factors of control. In Robert
Van Valin (ed.), Advances in Role and Reference Grammar, 167–196. Amster-
Declerck, Renaat. 1997. When-Clauses and Temporal Structure. London: Routledge.
Dedrick, John M. and Eugene H. Casad. 1999. Sonora Yaqui Language Structures.
Tucson, AZ: University of Arizona Press.
Diessel, Holger. 2008. Iconicity of sequence: A corpus-based analysis of the
positioning of temporal adverbial clauses in English. Cognitive Linguistics
19(3): 465–490.

Diessel, Holger. 2013. Adverbial subordination. In S. Luraghi and C. Parodi

(eds.), Bloomsbury Companion to Syntax, 341–354. London: Continuum.
Diessel, Holger and Katja Hetterle. 2011. Causal clauses: A cross-linguistic
investigation of their structure, meaning and use. In P. Siemund (ed.),
Linguistic Universals and Language Variation, 23–45. Berlin: Mouton.
Dixon, R. M. W. and Alexandra Aikhenvald. 2009. The Semantics of Clause
Linking. Oxford: Oxford University Press.
Foley, William A. 2010. Clause linkage and nexus in Papuan languages. In
Bril (ed.), 51–104.
Ford, Cecilia. 1993. Grammar in Interaction: Adverbial Clauses in American English
Conversations. Cambridge: Cambridge University Press.
Gast, Volker and Holger Diessel. 2012. Clause Linkage in Cross-Linguistic Perspec-
tive: Data-Driven Approaches to Cross-Clausal Syntax. Berlin: Mouton.
Gerardo-Tavira, Rebeca. 2018. Oraciones subordinadas temporales: orden, iconici-
dad y relaciones entre eventos. PhD dissertation, Universidad Nacional
Autónoma de México.
Givón, Talmy. 2001. Syntax: An Introduction. Amsterdam: John Benjamins.
Guerrero, Lilián. 2006. The Structure and Function on Yaqui Complementation.
PhD thesis, University at Buffalo (SUNY).
Guerrero, Lilián. 2009. On the semantic dimension of complementation. In
L. Guerrero, S. Ibáñez and V. Belloro. Studies in Role and Reference Grammar,
319–343. Mexico: UNAM.
Guerrero, Lilián. 2011. Clause linkage and purpose clauses in Southern Uto-
Aztecan Languages. In W. Nakamura (ed.), New Perspectives in RRG, 217–245.
Guerrero, Lilián. 2012. Más sobre controladores y pivotes: el caso de las
cláusulas de propósito. In R. Mairal Usón, L. Guerrero and C. González-
Vergara (eds.), El funcionalismo en la teoría lingüística: la Gramática del Papel y
la Referencia, 307–329. Madrid: AKAL.
Guerrero, Lilián. 2013. Controller–controllee relations in purposive con-
structions: A construction based account. In B. Nolan and E. Diedrichsen
(eds.), Linking Constructions into Functional Linguistics: The Role of Constructions
in Grammar, 1–22. Amsterdam, John Benjamins.
Guerrero, Lilián. 2017. On purpose, causal and reason clauses in Yaqui.
International Journal of American Linguistics 83: 679–718.
Guerrero, Lilián. 2019. Yaqui clauses and the Interclausal Relations Hier-
archy. In R. Kailuweit, L. Künkel and E. Staudinger (eds.), Applying and
Expanding Role and Reference Grammar. Germany: NIHIN.
Guerrero, Lilián, V. Belloro and Carmen Conti. 2017. Motivaciones en con-
flicto en la posición de adjuntos temporales de secuencia. Onomázein 36
(June): 98–121.
Guillaume, Antoine. 2008. A Grammar of Cavineña. Berlin: Mouton.
Hengeveld, Kees. 1998. Adverbial clauses in the languages of Europe. In J.
Van der Auwera (ed.), Adverbial Constructions in the Languages of Europe
335–420. Berlin: Mouton.
Hetterle, Katja. 2015. Adverbial Clauses in Cross-Linguistic Perspective. Berlin: Mouton.

Lehmann, Christian. 1988. Towards a typology of clause linkage. In J. Hai-

man and S. A. Thompson (eds.), Clause Combining in Grammar and Discourse,
Matić, Dejan, Rik van Gijn and Robert D. Van Valin, Jr. 2014. Information
structure and reference tracking in complex sentences. In R. van Gijn, J.
Hammond, D. Matić, S. van Putten and A. V. Galucio, Information Structure and
Reference Tracking in Complex Sentences, 1–44. Amsterdam: John Benjamins.
Matthiessen, Christian M. I. M. and Sandra A. Thompson. 1988. The structure
of discourse and ‘subordination’. In J. Haiman and S. A. Thompson (eds.),
Clause Combining in Grammar and Discourse, 275329. Amsterdam: John
Benjamins.
McDaniels, Todd. 2014. Rationale and purposive clauses in Comanche. Inter-
national Journal of American Linguistics (80): 69–97.
Pérez Quintero, María Jesús. 2002. Adverbial Subordination in English:
A Functional Approach. Amsterdam: Rodopi.
Schmidtke-Bode, K. 2009. A Typology of Purpose Clauses. Amsterdam: John
Benjamins.
Thompson, Sandra, Robert Longacre and Shin J. Hwang. 2007. Adverbial
clauses. In T. Shopen (ed.), Language Typology and Syntactic Description,
Vol. 2, 237–300. Cambridge: Cambridge University Press.
Vallejos, Rosa. 2014. Reference constraints and information structure man-
agement in Kokama purpose clauses: A typological novelty? International
Journal of American Linguistics 80(1): 39–67.
Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantic Interface.
Van Valin, Robert D., Jr. 2007. Recent developments in the Role and Reference
Grammar theory of clause linkage. Language and Linguistics 8(1): 71–93.
controllers. In L. Guerrero, S. Ibáñez and V. Belloro (eds.), Studies in Role and
Reference Grammar, 45–68. México: UNAM.
Van Valin, Robert D., Jr. 2021. Cosubordination. In R. Van Valin (ed.), Chal-
lenges at the Syntax–Semantics–Pragmatics Interface: A Role and Reference Gram-
mar Perspective, 241–254. Newcastle upon Tyne: Cambridge Scholars
Publishing.
Wash, Suzanne. 2001. Adverbial Clauses in Barbareño Chumash Narrative Dis-
course. PhD dissertation, University of California, Santa Barbara.
Notes
1 Previous works dealing with adverbial clauses in RRG include Bickel

(1993, 2003, 2010), Conti (2012), Van Valin (2007, 2009), and Guerrero
(2012, 2013, 2019). Some recent cross-linguistic studies dealing with

complex constructions have also emphasized unique aspects of the

juncture–nexus theory (Schmidtke-Bode 2009; Bril 2010; Gast and Diessel
2012; Matić et al. 2014; Hetterle 2015); see also Conti (2021) for a discus-
sion on cosubordination and other linkages on Spanish.
2 In the examples, coreferential and shared arguments are co-indexed, and
‘_’ signals an implicit argument; in non-English examples, the linked
unit is in brackets.
3 In While washing the car, Dana saw Chris, Dana may be understood as the
controller of the missing argument.
4 See Van Valin (2005) for a detailed description of the theory of linking in
simple and complex constructions.
5 Although operator sharing is an important feature, ‘there appear to be
cases of cosubordination in which operator sharing is possible but not
obligatory’ (see Bickel 2003, 2010; Van Valin 2007: 80; Foley 2010). For
instance, Bickel (2010: 53) found that in Belhare, Nepali, and several
other languages of South Asia, the scope of interrogative markers is
indeterminate: depending on the context, the sentence may be inter-
preted as having conjunct or disjunct scope. In some cases, these sen-
tences can be read as conjunctions or temporal relations. Van Valin
(2021) revisits Foley’s and Bickel’s concerns on the variable scope of the
IF operator and explains how the problematic examples do not question
the validity of cosubordination as a nexus relation, but they do highlight
the language-specific constraints at the clausal and sentence levels.
6 See also Van Valin (2005: 282–284) for discussion of IF operators in
reason, conditional and concessive clauses.
7 In the final reformulation, Van Valin uses only ‘and’, meaning that both
steps must be satisfied. If ‘or’ is also included in the linking, non-
subordinate core junctures in peripheral positions can be neatly cap-
tured, without additional modifications.
8 The logical structure in (25a′) tries to capture the notion of motion-cum-
purpose as modifying subevents. In Van Valin (2005), the formal repre-
sentation of a purposive relation is want’ (xi, [LS2. . . x. . .]) ^ DO (xi, [LS1. . .
xi . . .] e CAUSE [LS2. . . xi. . .]). In (25b), this logical structure captures the
idea that a volitional action is performed with the intention of another
event, and the fact that the main actor is directly involved in the
intended event (same-subject clauses). Since the betchi’ibo-clauses in
Yaqui are not implicative, the logical structure in (25c) does not contain
the CAUSE component. PURP in (25d–e) is a simplified version of the
logical structures in (25b–c).
9 For details on the analysis of control constructions in RRG, see §7.3.1 in
Van Valin (2005).
10 This is a highly marked nuclear juncture; see Van Valin and LaPolla
(1997: 671, fn. 19) for another example found in French involving
perception verbs.

15
Cleft Sentences and
Relative Clauses
Luis París
15.1 Introduction
Role and Reference Grammar (RRG) holds the view that a sentence is not just
a concatenation of words into a sequence but the incarnation of a set of
formal templates that are stored in syntax.* Syntactic templates have two
effects that are crucial for the understanding of the nature of linguistic
symbols. First, they are a key factor in determining the pattern followed by
the influx of information contributed by each word into the new symbol (i.e.
the phrase). Second, the phrase is an enclosure where certain relations occur
among its constituents that are shielded from some external influences.
Cleft sentences and relative clauses make this encapsulation of the phrase
evident. Sentence (1) illustrates a typical instance of a relative clause struc-
ture (hereafter, RC).
(1) The student who failed the test quit the program.
This sentence is an instance of an externally headed relative clause (EHRC),

which will be analysed in Section 15.2. It contains a relative clause (‘who failed
the test’) that modifies the head noun (i.e. ‘student’) of the referential phrase
(RP) – in which the relative clause is embedded. This head noun is the antece-
dent that fixes the reference of the relative pronoun (‘who’) in the relative
clause and, in doing so, it might also demand agreement on nominal features
like person, gender and number depending on the morphology of the lan-
guage. The head noun functions in (1) as the privileged syntactic argument
(PSA) of the main clause; however, relative clauses can modify any non-PSA
direct core argument and even adjuncts modulo Keenan and Comrie’s (1977)
Accessibility Hierarchy.1 In the same sense, the co-indexed wh-form can have
the function of any direct core argument or it can even be an adjunct in the
* This paper has benefited greatly from the careful reading, insightful questions and several suggestions by Delia Bentley, one
of the editors of this Handbook. Of course, I am responsible for all remaining errors.

592 LUIS PARÍS
relative clause (though see Section 15.2 for some relevant restrictions in
Malagasy, a Western Austronesian language spoken in Madagascar). Therefore,
the shared participant is not required to play any specific syntactic or seman-
tic role in either the matrix clause or the relative clause. This has important
consequences for the syntax of RCs. The relative clause is safeguarded from
any requirement or influence from the matrix clause by the RP in which it is
embedded. Operators like tense, grammatical aspect or negation affect each
clause (matrix and relative) independently. All the formal properties imposed
on the relative pronoun come from the head noun – and are of a referential
nature (person, gender and number) – or from the matrix verb of the relative
clause. These properties are captured by an analysis of relative clauses as RP
nuclear modifiers (Pavey 2004; Van Valin 2005: 220–223; 260–266; Van Valin
and LaPolla 1997: 497–499). This means that RCs, just like English attributive
adjectives, are elements in the periphery that describe either a property of the
referent or an event in which it participates. This semantic dependency is
reflected in the syntax and further justifies the nuclear periphery analysis: for
those languages with a rich nominal morphology, the noun imposes agree-
ment relations on the attributive element.
The semantics of RCs displays properties that illustrate the insufficiency of
truth conditional semantics to account for linguistic meaning. The meaning
of the relative clause and the head noun in (1) can be expressed by the
following simple sentence.
(2) The student failed the test (and quit the program).
Thus, the information about the world is the same in both cases. However,
the meanings are quite different because the composition of the respective
semantic representations is different. This will be captured later in this
chapter (see the representation in (10)) by embedding the logical structure
(LS) of the RC as an argument of the attributive LS, which is consistent with
the RC syntax, in that the relative clause is a modifier of the RP nucleus. In
addition to EHRCs, Section 15.2 will also discuss Internally Headed Relative
Clauses (IHRCs), free relatives and non-restrictive relatives. In IHRC, the head
noun is inside the embedded clause, which we know is embedded due to the
presence of a complementizer. The so-called ‘free relatives’ are not really
RCs, since they lack a head noun as well as a relative pronoun in many
languages. Finally, in non-restrictive RCs, the embedded clause is separated
from the head noun by a pause, and it does not fulfil the same pragmatic
function characteristic of relative clauses, namely, the identification of the
referent of the head noun.
This chapter also discusses cleft sentences, specifically, it-clefts as exempli-
fied by (3), which will be the focus of Section 15.3.
(3) It was John who asked the question.
The components of it-clefts are a matrix clause with a ‘dummy’ ‘it’ pronoun,
a copula clause and a clefted noun phrase (John in (3)), which is the

Cleft Sentences and Relative Clauses 593
antecedent that fixes the reference of either a relative pronoun or, if this is
absent, an argument or adjunct slot in the embedded clause, as in (4).
(4) It was in 1996 Mary had her baby!
In (3) the subordinated clause is introduced by a relative pronoun but it can

also be subordinated by a complementizer, as in (5).
(5) It was in 1996 that Mary had her baby!
The analysis offered in this chapter adopts a perspective that highlights

the interfaces that constitute the internal structure of clefts. From a truth-
conditional perspective, sentence (3) carries the same content as a simple
sentence like John asked the question. In fact, (3) describes a single event
whose description is contributed by the LS of the embedded verb (i.e. ask).
In this sense, the embedded clause provides the bulk of the semantic
content carried by the whole sentence, whereas the matrix clause contains
almost no truth-conditional information. This is the reason why this chap-
ter introduces the notion of clefts as a structure that materializes an
‘inverted’ linking. By ‘inverted linking’ I mean that the stronger semantic
clause, or the clause that provides the meaning for the whole sentence, is
syntactically dependent on a semantically weaker matrix clause (see
Section 15.3). This linking is not accidentally associated with a particular
information structure. In clefts, matrix clauses carry the pragmatic infor-
mation weight, since they contain the narrow focus of the sentence,
namely, the clefted noun (i.e. John in (3)). Further, this information is also
the reference value of the relative pronoun in the dependent clause.
Therefore, the matrix clause is semantically weaker, yet informatively
stronger, than the dependent clause, so that the syntax of clefts is clearly
driven by pragmatics. This conclusion is consistent with the view held in
RRG on the depth of the insertion of communicative needs in grammatical
structure. Information structure digs down into the core of the
grammatical engine.
There are strong similarities between RCs and clefts. The most salient one
is that both constructions involve an embedded clause that can be intro-
duced by a relative pronoun, which might be absent in some contexts. In
addition, both the embedded clause in clefts and those in RCs are presup-
posed. These resemblances motivate some proposals in which cleft clauses
are analysed as relative clauses. In contrast, the view pursued in this chap-
ter – in particular in Section 15.4 – focuses rather on the differences between
the two types of embedded clause. One of the properties of relative clauses as
defined here is that they are modifiers of the nominal nucleus of a noun
phrase. There is clear evidence that this is not the case with clefts. For
instance, as will be discussed in due course, in RCs, quantifiers in the matrix
clause range over the reference of the noun as restricted by the relative
clause, while in clefts they have scope over the meaning of the antecedent
noun without any restriction imposed by the embedded clause. This

594 LUIS PARÍS
evidence leads to a representation of clefts in terms of a clause-linkage

relation in which the embedded clause modifies the core of the matrix
clause. Section 15.4 will also explore fundamental pragmatic differences
between the two constructions. The embedded clause in an RC is a presup-
position that helps the listener to identify an otherwise unidentifiable
participant. In contrast, the activation status of the referent of the clefted
noun in clefts is irrelevant; what is crucial here instead is that this noun is
the focal element.
The chapter is organized as follows. Section 15.2 presents an analysis of
RCs: their syntax, semantics and syntax–semantics linking. Section 15.3
introduces cleft sentences and subsection 15.3.1 the relevant portion of
the linking algorithm that accounts specifically for clefts. Section 15.4 offers
a contrastive analysis of clefts and RCs. The chapter ends with Section 15.5,
which presents a short conclusion.
15.2 Relative Clauses
The ingredients of a typical English RC like (1), repeated in (6a) for ease of
exposition, are a clause introduced by a relative pronoun (i.e. who) co-indexed
with a head noun outside the RC and preceding it (i.e. student) and, finally, a
matrix clause, which the head noun belongs to. The RC is embedded within
the RP headed by the antecedent noun – student – as can be seen, for example,
by the fact that a proform can replace the entire RP (see 6b).
(6) a. The student who failed the test quit the program.
b. She quit the program.
The events described by the RC – the failing event – and by the matrix
clause – the quitting event – might not be connected beside the fact that
they share the same participant. A central relation within the RP is that of
obligatory co-indexation between the head noun and the relative pronoun.
As long as it belongs to the RP, there may be intervening material – all
within the RP that contains the relative clause – between the head noun and
the relative pronoun, as in (7).
(7) The student of the teacher who called you yesterday is sick.
However, the two nouns have the right features to function as antecedent of
the pronoun – they are both [þhuman] – and the interpretation is ambigu-
ous. Clearly, the preferred interpretation takes the closer antecedent (i.e.
teacher). In addition, if there is more available intervening material between
the head noun and the relative pronoun, the chances of acceptability of the
sentence are proportionally lowered.
The fact that there is no restriction on the function that the head noun
can have in the matrix clause is illustrated below. It can be a direct core
argument like in (6a) or it can even be an adverbial as in (8).

(8) I saw Mary at the bar where you met her.
In this case the relativizer is not a pronoun but a wh-form with an adverbial
function in the RC; in other languages, it could be any pro-adverbial form.2
In English as well as many other languages there are no restrictions on the
function of the co-indexed proform in the relative clause, as suggested by (8).
However, some languages pose specific requirements. As noted by Keenan
(1976), Malagasy allows only the co-indexed element to be the subject of the
relative clause (9).
(9) a. Na-hita ny vehivavy [(izay) nan-asa ny zaza] Rakoto.

pfv.atv-see det woman comp pfv.atv-wash det child Rakoto
‘Rakoto saw the woman that washed the child.’3
*‘Rakoto saw the woman that the child washed.’
b. Na-hita ny zaza (izay) nan-asa ny vehivavy Rakoto.
pfv.atv-see det child comp pfv.atv-wash det woman Rakoto
‘Rakoto saw the child that washed the woman.’
*‘Rakoto saw the child that the woman washed.’
c. Na-hita ny zaza (izay) sas-an’ ny vehivavy Rakoto.
pfv.atv-see det child comp wash-pass det woman Rakoto
‘Rakoto saw the child that was washed by the woman.’
RCs require a matrix clause with an available antecedent noun that

heads an RP that contains an embedded clause via a proform co-indexed
with the antecedent noun. The lexical category of the antecedent is fixed: it
can only be a noun. This is easily relatable to the communicative function
generally attributed to (restrictive) RCs, namely that they help identify the
referent of the antecedent noun by relating it to some already
known event.
The syntactic representation of the EHRC in (1) is given in Figure 15.1. Just
like any adjectival modifier, the relative clause is in the Periphery of the
nominal nucleus of the Referential Phrase.
The semantic representation expresses the contribution of the RC as an
attributive modifier (i.e. be′ (. . .)) of the head noun. This predication is
inserted into the matrix LS slot that corresponds to the first argument,
that is, the first argument of quit′, which should be the position occupied
by the head noun. The participant referred to by the head noun student is
shared with the predicative LS (be′(student, . . .). In short, (10) represents
that the semantics of the RC fills in the argument slot in the main LS that
corresponds to the head noun. In particular, this is not a co-indexation
relation, but a position that belongs to two LSs (the one that corresponds to
quit′ and the one that corresponds to be′). The slot of this shared partici-
pant is shown in (10) by a thick broken underline (Van Valin 2012). In
addition, this position is co-indexed with the argument position filled with
the wh-word in the relative clause LS.4 The superscript ‘¬i’ refers to the
activation status of the participant; in particular, the head noun of the RC
is unidentifiable, which mean that it is neither active or accessible nor

596 LUIS PARÍS
SENTENCE
CLAUSE
CORE
RP NUC RP
CORE PRED
NUCN PERIPHERY
CLAUSE
PrCS CORE
RP NUC RP
PRED
The student who failed the test quit the program

Figure 15.1 Externally headed relative clauses
inactive, it is unidentifiable and either anchored or non-anchored (Chafe

1987; Pavey 2004: 5).
(10) INGR quit′ ([be′ (student1¬i, [INGR fail′ (who1, test)])], program)
----------
The regular syntax-to-semantics linking algorithm applies to relate
Figure 15.1 to (10) but, in addition, we need the following further restric-
tions (Van Valin 2012: 57).
(11) a. Retrieve from the lexicon an attributive LS (i.e. be′ (x, [pred′])) and
substitute the LS of the verb in the relative clause (i.e. [INGR fail′ (x, y)]) for
the second argument of the attributive LS (i.e. [pred′]).
b. Co-index the first argument in the attributive LS with either the unlinked
argument position in the relative clause LS, or, if there is a relative
pronoun, to the argument position linked to the relative pronoun.
c. Insert the attributive LS into the argument position in the matrix LS
occupied by the head noun, replacing the variable in the first argument
position in the attributive LS with the head noun.
There is still the question whether this structure complies with the
Completeness Constraint in cases like (12), where there is no overt relative
pronoun. The absence of an overt relative pronoun is possible in all cases
except in those where the pronoun is linked to the argument in the LS that
should be the PSA/subject.
(12) The student I saw in the library was not screaming.
In this case, the undergoer of see is not realized within the embedded
clause. This requires the reduction of a slot in the verb core, although the

structural conditions for this reduction are entirely different from those
that characterize control structures (Van Valin 2005: 255). Control
involves a non-subordination nexus at a core juncture, in which the
relevant core is a semantic argument of the matrix core. Quite differ-
ently, the relative clause in RC is a subordinated optional modifier of a
noun nucleus.
The answer to this problem is to resort to the formulation of the Com-
pleteness Constraint in Van Valin (2005: 233). Under this principle, the
argument of an LS that is not syntactically realized within the clause can
be satisfied anywhere within the sentence that contains that clause. In this
way Completeness is respected. Still, as is the case with control structures,
we need to determine the specific structural conditions that license this slot
reduction in RCs. The following is an attempt to capture the relevant
restrictions.
(13) The syntactic slots of the core of a finite clause may be reduced by 1 if the
argument to which the missing slot should have been linked is fulfilled
by a variable co-indexed with an antecedent noun that is the nucleus
modified by the phrase containing the empty slot.
In this way we identify the specific condition that allows the optional
reduction of an argument slot in some English EHRCs.
Internally headed relative clauses are characterized by the head noun
being part of the relative clause. In particular, the head noun is expressed
in a position within the relative clause, while it is omitted in the matrix
clause. This type of RC is found in Bambara, an SOV language.
(14) Bambara (Mande, Africa; Bird 1968, cited in Van Valin 2012)
[Ne ye so min ye] tye ye san.
1sg pst horse rel see man pst buy
‘The man bought the horse that I saw.’
The relevant question is how the hearer recovers the semantic function of
the co-indexed participant in the matrix LS. Upon hearing (14), a transitive
syntactic template is retrieved because of the verb buy. There is a relativizer
which indicates the presence of an embedded clause headed by the verb see,
which contains two arguments: ne (‘I’) and so (‘horse’). The word order of the
language (SOV) indicates that the first one is the Actor (and, hence, the PSA)
while the second is the Undergoer and, thus, the non-PSA macrorole argu-
ment. In turn, the matrix clause nucleus is buy, which requires two argu-
ments, although there is only one: tye (‘man’). The fact that the verb buy
selects human actors is the only cue that indicates, first, that man is the
Actor and, second, that there is a missing Undergoer argument, which has to
be found in the relative clause. There are two candidates, one is I and the
other is horse. World knowledge indicates that people are less likely to be
sold than horses and this might motivate the choice of horse in this case. In
other cases, the structure may be disambiguated on the basis of contextual

598 LUIS PARÍS
SENTENCE
CLAUSE
CORE
RP RP NUC
CLAUSE
CORE
RP RP NUC
Neye so min ye tye ye san.

Actor Undergoer
Actor
LEXICON see′(x, y) Undergoer
(15a) [do′ (v, Ø)] CAUSE [BECOME have′ (v, w)] LEXICON
[be′ (x, [pred′])] [be′ (xi , [see′ (ne, soi)])]
co-indexing (15b)
RC LS substitution (15c)
[do′ (tye, Ø)] CAUSE [BECOME have′ (tye, [be′ (xi, [see′ (ne, soi)])])]
Figure 15.2 Internally headed relative clauses (from Van Valin 2012)
clues. Taking into account the preceding considerations, the rules governing
linking from syntax to semantics in IHRCs are the following (Van Valin
2012: 60):
(15) a. Retrieve from the lexicon an attributive LS and substitute the LS of the
verb in the relative clause for the second argument.
b. Co-index the first argument in the attributive LS with the argument in the
relative clause LS identified as the head noun.
c. Insert the attributive LS into the open argument position in the matrix LS.
The constituent structure of this construction, shown in Figure 15.2, is

extremely interesting. One of the salient features is that it allows the
antecedent of a co-indexation relation to be more embedded than the co-
indexed variable. Typically, the antecedent precedes the co-indexed element
in the semantic representation and the antecedent is either equally or less
syntactically embedded than the co-indexed element. However, this is not
the case with IHRCs.
In IHRC, the Completeness Constraint is satisfied since the argument
linked to a reduced slot is satisfied somewhere else within the sentence.
However, it is worth pointing out the structural difference between

reducing an argument slot in the matrix core and reducing it in an embed-

ded clause. In particular, binding a variable with an argument nested more
deeply in the LS is not at all frequent in languages like English, if possible at
all, but this is exactly the case with [be′ (xi, [see′ (ne, soi)])], where the lexical
element so is an argument of an LS contained within another LS with a
variable bound to the embedded lexical argument. This relative clause is not
a modifier of an RP, but rather the relative clause and the co-indexed
element are arguments of the same clause.
Another issue raised by IHRCs concerns co-indexation. Typically, obliga-
tory co-indexation like the one exhibited in RCs is governed by structural
restrictions. In the case of obligatory reflexives, the antecedent must hold a
semantic role that is higher on the macrorole hierarchy than the one held
by the co-indexed element. Ultimately, this principle is motivated by the
semantic role hierarchy, and therefore, it is independently motivated (Van
Valin 2005: 61). Control structures are another example: the controller is an
argument of the matrix verb and the controllee is an argument in the LS
embedded in the matrix verb LS.
In EHRCs the antecedent is higher than the proform both syntactically
and semantically. Syntactically, the antecedent is an argument of the matrix
clause that contains the clause linked to the LS with the co-indexed proform.
Semantically, the higher status of the antecedent does not need to match a
higher ranking on the Actor-Undergoer Hierarchy, since they are mapped to
macroroles within different LSs.
(16) The dog bit the boy that killed the spider.
The antecedent boy is an Undergoer in the matrix clause. Yet, the co-indexed
relative pronoun is an Actor in the embedded clause. Therefore, the higher
status of the antecedent bears on the LS that contains it in relation to the LS
that includes the proform.
The Bambara IHRC is quite different in this regard. The antecedent is
within a clause that is at the same syntactic level as the reduced slot. In
turn, the semantics contains the bounded variable in the LS of the depend-
ent verb, which is nested within the attributive structure that contains the
antecedent as an argument. This contrast – plus the fact that the relative
clause precedes the matrix clause – strongly suggests that the co-indexing
relation in Bambara IHRCs is driven by pragmatics and world knowledge, as
we argued above.5
As for non-restrictive relative clauses, these are clearly different from
restrictive ones, in that there is a pause separating the relative clause from
its antecedent. In addition, the attribution does not serve the purpose of
helping the hearer to identify the referent. Indeed, as can be seen in (17), the
antecedent can be a proper noun. Consequently, the information it provides
does not need to be presupposed. These properties are well taken care of by
the analysis in Figure 15.3, in which the relative clause is a peripheral
modifier of the RP (Van Valin 2005: 222).

600 LUIS PARÍS
RP Periphery
CLAUSE
N PrCs CORE
RP NUC RP
CORE
NUC
Chris, who loves soccer

Figure 15.3 Non-restrictive relative clauses
(17) Chris, who loves soccer, flew to Barcelona.
Finally, we should briefly analyse the so-called ‘headless relative clauses’

or ‘free relative clauses’ as exemplified in (18) and (19). The embedded
clauses in those sentences (what John said and los que rompieron el libro ‘those
who tore the book’) are introduced by wh-words (what and los que) which
satisfy argument positions in the matrix LS. The wh-words are in turn
modified by an embedded clause.
(18) What John said surprised Sandy.
(19) Los que rompieron el libro corrieron a sus casas

those.m.pl that tear.pst.3pl the-m.sg book run. pst.3pl to their house.pl
‘Those who tore the book apart ran to their houses.’
From a purely semantic standpoint, these examples are like relative

clauses, since the argument position in the matrix clause is filled in by
the entity referred by the wh-word. This entity is in turn modified by the
LS of the embedded verb. In particular, it is not the case that the LS of the
embedded clause is an argument of the matrix LS, but rather it is part of
the information that fills in that argument position. This meaning is
captured by (20), an LS that is structurally identical to the one assigned
to the relative clauses previously analysed. There is a crucial difference,
though. In relative clauses the underlined argument is filled in by a full
noun that is the head of the relative clause and functions as the antece-
dent of the relative pronoun. In contrast, in (20) the position is occupied
by a pronoun and the typical slot of the relative pronoun in the attribu-
tive LS is filled by a variable co-indexed with the pronoun. It is clear from
(20) that the entity that caused the surprise is the referent of what, and
the LS that corresponds to the embedded clause helps to restrict the
possible referents of that pronoun, which is just what plain restrictive
relative clauses do. In addition, the description in (20) derives the funda-
mental property of the pronoun in (18) and (19), namely that they do not

have antecedents and are thus free to any (contextually appropriate)

interpretation.
(20) [do′(whati [be′[do′ (John, say′(John, xi)])]) CAUSE [BECOME surprised′(Sandy)]]

-------
Is there any formal correlate of this semantics? The most prominent one
is that the nominative inflexion on the main verb in (19) is plural and is
being controlled by the pronoun los que ‘those that’. Crucially, the inflex-
ion is not determined by the whole embedded clause los que rompieron el
libro but only by the pronoun. In addition, the nominative case of the
pronoun is also determined by the matrix clause. In a sentence where the
pronoun plays a non-PSA role in the matrix clause, it has to be expressed
with an accusative/dative marker (i.e. a ‘to’) like in Pedro reemplazó a los
que llegaron tarde ‘Peter replaced those who arrived late’. Had the embed-
ded clause determined the shape of the pronoun, it would have been just
los que (nominative) not a los que (accusative). This is also true in cases
where the pronoun has an Actor role in the matrix clause and an Under-
goer role in the embedded clause (for example, El que invitaste llegó tarde
‘The one you invited arrived late’). In such a case the pronoun should be
realized as a nominative form consistent with its meaning in the matrix
clause.6 In contrast, the relative pronoun in plain relatives takes the
shape that is consistent with the semantics it takes from the embedded
clause. For example, in El niño al que le regalé el juguete sonrió ‘The boy
whom I gave the toy smiled’: the pronoun expresses the Beneficiary of the
verb regalar ‘to give’ and the Actor of the main verb sonreír ‘to smile’, but
its formal realization is consistent only with its Beneficiary role in the
relative clause.7
The generative tradition assumes either a ‘matching’ or a ‘raising’ analy-
sis of free relatives which subsumes them within plain relative clauses
(Cinque 2015). Both analyses assume the presence of an empty position
outside the relative clause (the position of the head/antecedent), which is
the landing site for the pronoun of free relatives. One of the problems of
subsuming the free relatives to plain relatives is that they do not behave in
the same way in the so-called ‘extraction’ phenomena. The antecedent of
the plain relative El discurso que diste sorprendió a Sandy ‘The talk you gave
surprised Sandy’ can be replaced by a wh-word as in ¿Qué que diste sorprendió
a Sandy? ‘What that you gave surprised Sandy?’. However, we cannot ask for
the identity of the agent of the matrix clause in (19) as in *¿Quiénes
rompieron el libro corrieron a sus casas? ‘*Who tore the book apart ran to their
houses?’. The sentences in (18) or (19) cannot be reduced to regular relative
clauses by claiming an empty position for the antecedent that is filled in by
the pronoun coming from the relative clause. They need a different and
specific treatment.
The evidence presented so far suggests, albeit not conclusively, that the
pronouns in (18) and (19) are not relative pronouns. They are more properly
analysed as arguments of their matrix clauses, as in Figure 15.4. We can call

602 LUIS PARÍS
SENTENCE
CLAUSE
CORE
RP Periphery NUC Periphery
CLAUSE V PP
CORE
NUC RP
Los que rompieron el libro corrieron a sus casas

‘Those who tore the book ran to their houses’
Figure 15.4 Pronounless relative clauses
the embedded clause ‘pronounless relative clause’ in order, first, to empha-

size the opposition with a headless relative clause analysis and, second, to
point to the exceptional nature of this ‘relative clause’. In sum, this is a
construction of its own that cannot be understood by just applying an old
label to it. The linking algorithm has to incorporate a statement that
guarantees that the interpretation of the non-syntactically realized argu-
ment of the verb in the embedded clause is driven by the co-indexation with
the pronoun in the matrix clause.
15.3 Cleft Sentences
Cleft sentences are as complex as they are interesting and, similarly to RCs,
they involve a matrix and a subordinate clause. There are different subtypes
of cleft, but in this chapter the focus will be on ‘it-clefts’, as represented by
sentence (21). This example includes a cleft clause (who blamed the bus driver)
introduced by a wh-pronoun that obtains its reference from the antecedent
noun (teacher) in the matrix clause, also called ‘clefted noun’. This matrix
clause consists of a copula with dummy it.
(21) It was the teacher who blamed the bus driver.
The wh-form that introduces the cleft clause can fulfil any direct core
argument position. In fact, it can even express adjuncts, as can be seen
in (22).
(22) It was in Buffalo where/that Shelly met her husband.
Clefts have properties that illustrate the complexity of the grammatical

interfaces in natural languages. The same semantic information carried by
(21) and (22) can be expressed by the simple sentences in (23) and, respect-
ively, (24) without any loss in the description of the relevant events.

(23) The teacher blamed the bus driver.
(24) Shelly met her husband in Buffalo.
The unmarked expression of a proposition is a simple sentence, but clefts

involve a bi-clausal structure with an unequal distribution of the content.
The semantic weight of (23) and (24) is conveyed by their respective verb LSs.
Yet, in clefts like (21) or (22), these content verbs are expressed in embedded
clauses. In some sense, this is an ‘inverted’ linking, in that the semantic
asymmetry by which one clause is richer in content than the other is not
reflected by the syntactic asymmetry. The dependent clause expresses the
semantically richer content, while the matrix clause conveys a weaker
meaning. This ‘inverted’ syntax–semantics interface is pragmatically driven.
It will be shown later that, while being semantically weaker, the matrix
clause expresses the element that bears the focus of the sentence. Further-
more, the syntactic realization of the clefted noun in the matrix clause is
determined by the semantic role this participant plays in the embedded
cleft clause, as clearly shown by (25). In particular, the PP to the clerk needed
to introduce this participant into the matrix clause is motivated by the
syntax–semantics interface of the embedded clause. This participant is the
third argument –Beneficiary – of a three-place predicate (give), therefore, it is
a non-macrorole argument that can thus be expressed by a to-PP. There is no
motivation for the formal properties of this argument (i.e. it being a prepos-
itional phrase) in the matrix clause.
(25) It was to the clerk that I gave my application.
Clefts are a primary example of a syntactic structure that is not entirely

semantically motivated. Indeed, we can speak metaphorically of a syntax–
semantics interface that is twisted for the expression of a particular kind of
information structure. This is a point that might contribute to our under-
standing of linguistic systems. At every moment speakers are faced with the
daunting task of capturing new situations and this means that every context
requires to some degree novel linguistic responses. The lexicon and the rules
that regulate the syntax–semantics interface are highly structured and, yet,
clefts show that they are flexible enough to accommodate to the demands of
a particular speech situation.
We can think of the communicative situation as the encounter of two
vectors coming from opposite directions in a sort of ‘pincer effect’: a struc-
turing force that holds the pieces together and the need to adapt to the
particular demands of a context. Clefts show that the supposedly more rigid
part of the system, syntax, has built-in templates specifically designed to
serve information structure, one of the subsystems that deals with adjust-
ment to the dynamics of contexts. There is no part of grammar that is
immune to the demands of communication.
Clefts allow speakers to highlight the particular insertion of the content
contributed by an isolated element – the clefted constituent – into the

604 LUIS PARÍS
incomplete content of the cleft clause. The single proposition has been
divided into an open proposition – a representation that comes with an
open slot or variable – expressed by the clefted clause and a matrix clause
that provides the value for that variable. The cleft clause is part of the
presupposition and, hence, outside the potential focus domain in a strong
sense: neither its internal constituents taken individually nor the entire
clause as a unit can be the focus. Example (21), repeated in (26a) for conveni-
ence, cannot be a felicitous answer to the question What happened?. The
potential focus domain only has scope over the matrix clause as confirmed
by focus sensitive-particles. The negative operator in (26b) ranges only over
the matrix clause and, consequently, on the co-indexation of the teacher as
the Actor of the embedded clause.
(26) a. It was the teacher who blamed the bus driver

b. It wasn’t the teacher who blamed the bus driver.
Further, a yes/no question can only ask for the clefted RP. Was it the teacher
who blamed the bus driver? can be answered by (26a) whereas Was it the bus
driver who was blamed by the teacher cannot. The appropriate context for an it-
cleft sentence is a contrastive one; it had been asserted that somebody else
other than the teacher (for example, a student) had accused the bus driver.
This contrastive focus typically conveys an exhaustive reading of the clefted
RP; namely, it was the teacher and nobody else who accused the bus driver
(Kiss 1998).
Languages like French have constraints preventing PSAs – subjects – to be
part of the actual focus domain in a sentence focus structure (SF). Since
French also has rigid SVO word order, preverbal elements cannot normally
be in focus. In consequence, French might use clefts as in (27) or presenta-
tional clefts as in (28) as pragmatically felicitous in a context that demands
SF (Lambrecht 1994: 226, 2000: 653; Van Valin 1999: 519).
(27) J’ai ma voiture qui est en panne.

I have.prs.1sg my car that be.prs.3sg in breakdown
‘My car broke down.’
(28) Y a Jean qui a téléphoné.

there have.pst.3sg Jean who have. pst.3sg call.ptcp
‘Jean called.’
These sentences can be a felicitous answer to a What happened? type of

question. The rather fixed French word order and the pragmatic restriction
of subjects as topics motivates the use of clefts.
In RRG the presence of a bi-clausal structure like the it-cleft raises the
issue of nexus and juncture linkage. ‘Nexus’ refers to the nature of the
dependency between the two clauses, specifically, whether one entirely
depends on the other (subordination) or they are independent of each
other (coordination) or, finally, they are co-dependent (cosubordination).
In addition, the theory of ‘juncture’ individuates the layer in which the

combination occurs: nuclear, core or clausal (see Chapter 13). Since clefts
contain an obligatory co-indexation relation among members of different
clauses, the corresponding juncture is at the core layer. Regarding nexus,
we can rule out daughter subordination since the subordinate clause does
not fill an argument slot in the matrix clause and the co-indexation rela-
tion might involve adjuncts in the embedded clause. Furthermore, the
embedded verb inflexion agrees with the wh-form and the clefted element
in the matrix clause does not control agreement. This can be seen in
Spanish where the antecedent noun is a first-person pronoun while the
embedded verb agrees with the third person that corresponds to the
embedded pronoun (Pavey 2004: 29).
(29) Fui yo el que chocó el auto.

be.pst.1sg I the.m.sg that crash.pst.3sg the.m.sg car
‘It was me who crashed the car.’
This should be another piece of evidence that supports a core peripheral

subordination in a core juncture analysis. In the RRG account, the nucleus
of a clause with a copula is the attributive expression, in this case, the RP.
The embedded clause is a modifier of a core with a nominal nucleus. In this
way the similarity between it-clefts and RCs is also captured (see
Figure 15.5).
The embedded clause in it-clefts is a peripheral modifier at the core layer
and, hence, cannot be part of the focus. There are some exceptions to this,
but they are other subtypes of clefts. For instance, there-clefts like the one in
(30) might be appropriate answers to What happened?
(30) There was one man that kept interrupting.

(Huddleston 1984: 489; Pavey 2004: 64)
This is an existential cleft with a lightly informative antecedent noun; the

cleft clause might be part of the focus (Bentley et al. 2015). Another example
is offered by it-clefts with ‘this’ or ‘that’ in place of the ‘it’ pronoun, like (31).
The information in the matrix clause retrieves already known information
SENTENCE
CLAUSE
CORE PERIPHERY
RP NUC CLAUSE
PRED PrCS CORE
AUX RP NUC RP
It was the teacher who blamed the bus driver.

Figure 15.5 Cleft sentence

606 LUIS PARÍS
so that the assertion falls directly on the identity relation between the
copula and the cleft sentence, which is the focus of the assertion.8
(31) a. That’s the reason I don’t want to go to Miami!

b. Yeah. Wasn’t that somewhere in Southern Florida where they
thought those people got AIDS from bug bites. . .
(Hedberg 2000: 902; Pavey 2004: 32)
Spanish does not have It-clefts but it does have a construction that is
structurally similar to the English one. It places the focus on the cleft
constituent in a clause with a copula followed by an embedded clause
introduced by a relative pronoun.
(32) Fueron los niños los que pidieron leche.

Be.pst.3pl the.pl child. pl the.pl that ask.pst.3pl milk
‘It was the children who asked for milk.’
A difference with English is that in the Spanish cleft, the clefted noun might
be the subject of the matrix clause – as in (32) – since it controls the
agreement with the verb inflexion (i.e. the copula). Hence, the absence of
an ‘it’ pronoun cannot be attributed to the Spanish ‘pro-drop’ status. Despite
the variety of relative clauses existing in Spanish, here the relative pronoun
has to be of a certain kind – that is, lo que (‘the.m.sg that’), la que (‘the.f.sg
that’), los que (‘the.pl that’) and so on. As in English, the morphosyntactic
realization of the antecedent noun is determined by the meaning of the
embedded verb. In (33) the subject is impersonal and the antecedent is the
nominal element of a non-predicative PP that marks indirect objects.
(33) Fue a los niños a los

be.pst.3sg to the.pl child.m.pl to the.m.pl
que les dio la leche.
that them give.pst.3sg the.f.sg milk
‘It was to the children to whom he gave the milk.’
Furthermore, if the co-indexed argument plays the role of an adjunct in the

cleft clause, it should be preceded by the relevant predicative preposition, as
shown in (34).
(34) Fue en el baño donde encontré la llave.

Be.pst.3sg in the.m.sg bathroom where find.pst.1sg the.f.sg key
‘It was in the bathroom where I found the key.’
In spite of being syntactically embedded, the verb LS of the cleft clause

determines the morphosyntactic realization of the matrix participant. This
is a key property that justifies a different analysis of RCs and clefts. The
syntax–semantics linking in clefts displays properties of an ‘inverted’ inter-
face. Why is it, then, that the cleft clause is not represented as the matrix
clause? The presence of a subordinator – either a relative pronoun or a
complementizer – is the crucial piece of evidence for the embedding of
the cleft clause. In addition, an antecedent that controls the reference of a

pronominal expression typically occupies the syntactic dominant position,

which in this case is the matrix clause.
15.3.1 Linking Cleft Sentences

I shall follow Pavey (2004: 215) in assuming that the LS that corresponds to it-
clefts is based on the be′ predicate structure that expresses the meaning of
clauses with a copula, as shown in (35). This predicate captures the specifica-
tional nature of the predication, one of the meanings that can be expressed
by the copula in English and Spanish, the one that corresponds to clefts. In
particular, the specificational meaning highlights the ‘exhaustive reading’
in the sense that it was the teacher and nobody else who blamed the bus
driver.9
(35) be′([INGR blame′(who1, bus driver)], teacher1)

----------
There is an important assumption underlying (35) which is that LSs are
not merely registering propositional contents (interpreted either as a
cognitive state or as truth-conditional content). In fact, if this were the
case, this LS should also correspond to The teacher blamed the bus driver
since, as I pointed out earlier, this sentence and the one in (21), (26a) have
the same propositional value (see also Lambrecht 1994: 22). On the con-
trary, it is assumed that LSs should also be sensitive to the way symbols
are combined which, in addition, highlights the systematicity of the
syntax–semantics–pragmatics interface. The salient property of clefts
rests on the satisfaction of communicative needs. The LSs of clefts should
be consistent with the fact that a speaker is interchanging information
with an addressee who is in a different cognitive state. It-clefts are nested
in a communicative context where one interlocutor had asserted that an
individual ‘x’ was the Actor of a blaming event. The speaker utters the it-
cleft to contradict such a statement and offer a different identity for that
participant.10
The syntax-to-semantics linking proceeds as follows. The clause with the
copula demands that we retrieve a specificational structure and the LS of the
verb in the embedded clause. Since it has no recoverable referential value,
the argument slots of the specificational LS are filled in by the LS of the
embedded clause and the full RP. This RP is the specificational element;
hence, it has to fill in the second argument position. Then, we proceed to co-
index the wh-pronoun in the embedded verb LS with the specificational RP. If
there is a complementizer, then, the co-indexation involves the unspecified
argument slot of the embedded verb LS. Finally, the bus driver is an RP in
postverbal position. The Undergoer is, in turn, linked to the lowest-ranked
argument in the verb LS. The highest-ranked argument is linked to the
remaining wh-word or, if there is no available wh-word, it remains
syntactically unexpressed.

608 LUIS PARÍS
15.4 Contrasting Clefts and Relative Clauses
There is a sense in which the central meanings of it-clefts and RCs are
equivalent in terms of truth conditions. Specifically, in the sentence The
student that failed the test quit the program, the truth-conditional content of the
RC can also be expressed by a simple sentence (The student failed the test) and
this exact content can be expressed by an it-cleft (It was the student who failed
the test). The semantic difference in truth-conditional terms is reduced to the
minimum; it-clefts add to the interpretation of the relevant noun the fur-
ther assumption that the student was the only member who performed the
action (exhaustive reading). The bulk of the semantic contrast lies in the way
the semantic representation is assembled and these different assemblages
are the encodings of different communicative functions. Speakers use RCs to
provide further information that will help hearers to identify the referent of
a head noun. It-clefts, instead, highlight the identity of one participant with
a particular role in one event.
RRG captures this divergence through an independently motivated dis-
tinction. It is assumed that the be′ predicate includes different kinds of
predications (Van Valin 2005: 48), even though they might all be expressed
via a copula verb in a particular language. Two of them are relevant here;
one is attributive predication and the other is specification (Declerck 1988: 47;
Pavey 2004: 29), represented in (36) and (37), respectively. The former states a
predication relation between a predicate and a referential expression. It
subsumes attributive adjectives and, thus, the relation between relative
clauses and their head nouns. In contrast, the specification relation co-
indexes two referential expressions; one might be a constant (i.e. Chris) and
the other is a definite description (the winner), a representation that allows us
to identify a unique individual in the world.
(36) a. Pat is tall

a′. be′(Pat, [tall′])
(37) a. Chris is the winner

a′. be′(Chris, [the winner])
For the purpose of representing it-clefts, the definite description of struc-

tures like (37a′) is replaced by the clefted noun. The LS that captures the
meaning of the sentence is thus shown in (38).
(38) be′([INGR blame′ (who1, bus driver)], teacher1).
----------
It is the head noun that provides the specifying information and, hence, it
plays the role of the predicative element. Indeed, it is predicated of the
embedded logical structure (i.e. ‘. . . blame′ . . .’) which constitutes the mean-
ing of the embedded clause. The element that receives this predication is the
co-indexed argument in the embedded verb LS (i.e. the wh-form).
In contrast, the matrix verb of the relative clause in RC provides the LS for
the whole sentence, as can be seen in (10), which is repeated below in (39).

The predicative structure (i.e. be′) that corresponds to the semantic relation
between the head noun and the relative clause is inserted into one of the
matrix LS argument slots (i.e. the first argument of quit′).The participant
that fills in a semantic slot is literally shared with the predicative LS. This is
not a typical co-indexation relation but a unique slot that belongs to two LSs,
which is shown with a thick broken underline, as was pointed out above. In
addition, this position is co-indexed with the argument slot filled by the wh-
word in the relative clause LS.
(39) INGR quit′([be′ (student1, [INGR pass′(who1, test)])], program)

----------
The site of the syntactic contrast resides in the linkage of the embedded
clause with the matrix clause. It-clefts involve a direct relation between two
clauses, while the relative clause in RC acts as a modifier of the head noun
and it is embedded into an RP in RC. This difference is attested, for example,
by the different scope of the universal quantifier in (40) and (41) (Davidse
2000: 1114, cited in Pavey 2004: 193).
(40) It was all the passengers who had committed the murder.
(41) All the students who attended will receive a bonus point.
In (40) the quantifier has scope over only the clefted noun, and the co-
indexed pronoun in the cleft clause inherits this interpretation. The full
RP denotes the set of ‘all passengers’ and all of them committed the murder.
In other words, the cleft clause does not affect the interpretation of the
noun phrase in the matrix clause. By contrast, in (41), the quantifier has
scope over the RP the students, which has already been restricted by the
relative clause. Therefore, the relevant set does not include all the (context-
ually possible) students but only the subset of them that attended. Indeed, it
is implicated that some students did not attend the class. The relative clause
restricts the set of the noun to a subset and it is only to this subset that the
universal quantifier applies. These properties can be derived naturally by
representing relative clauses within the RP whereas cleft sentences are
outside the RP, as shown in Figure 15.6.
With respect to information structure, neither the relative clause in RC
nor the cleft clause in it-clefts are part of the assertion, but are instead
presupposed. They both carry a proposition with a variable, namely, an open
proposition. However, these embedded clauses differ in subtle ways. The
relative clause helps the interlocutor to identify the reference of the head
RP, which is necessarily a non-identifiable participant. In contrast, the
Det Noun embedded clause Det Noun embedded clause

It-clefts Relative Clause
Figure 15.6 The contrast between cleft sentences and relative clauses

610 LUIS PARÍS
reference of the clefted noun could be identifiable or not, but it receives an

exhaustive reading and, crucially, its pragmatic function as focus is to
saturate the open proposition introduced by the embedded sentence. The
cleft clause provides the event in which this entity was involved. Under-
standing focus in Lambrecht’s sense, the embedded clause in clefts carries
information that is necessarily part of the focus. In the example we have
been analysing, this is the semantic role ‘Actor’ in the blaming event of the
teacher (clefted noun). The teacher is the focal element but the focus is not
only the identity of the teacher but also its role (Actor) in the event described
by the embedded clause. Let’s call the kind of open propositions carried by
embedded clauses in clefts ‘operative presuppositions’ since they carry infor-
mation that is necessarily in focus. The embedded clause in RCs also carries
an open proposition, but it does not function in the same way in infor-
mation structure. Its information helps in the identification of a participant
that might not be in focus at all. In sum, RCs are related to the issue of
identifying a referent whereas clefts are centred on focus structure and,
consequently, their respective embedded open propositions have different
functions: one is an ‘operative presupposition’, whereas the other is just a
plain presupposition. Their syntactic templates are consistent with their
different informative roles since the relative clause is embedded into the
RP, whereas the cleft clause is a modifier of the matrix core and it does not
play any role within the RP.
In some sense, any sentence with a pronominal form is associated with an
open proposition. The pragmatic notion of open proposition that is relevant
here is that of an open proposition that is specified by the focus. The
embedded clauses of both constructions are open propositions, but only
the cleft clause is an open proposition in a pragmatically relevant way,
namely, it is the presupposition specified by the element in focus.
One condition that makes it possible for the relative clause in RC to be
presupposed, even though it carries specifying information, is to be found in
its syntactic layout. While the cleft clause is outside the clefted RP, the
relative clause is a modifier within the RP in RC. The embedding of the
relative clause in the RP leaves it out of the potential focus domain in a weak
sense, that is, it can be the focus as a whole, but its internal constituents
cannot be in focus on their own. This is shown in (42a), where the wh-form
targets the whole RP in (42b).
(42) a. Which student quit the program?

b. The one that failed the test.
Another contrasting behaviour related to the inherent pragmatic func-

tions of the constructions is that the head noun in it-clefts can be a proper
noun whereas this is not the case with RCs, as shown in (43) and (44),
respectively.
(43) It was Peter who sold the house.

(44) *Peter who sold the house came to see you.
In sum, the relative clauses in RC can be the focus as a whole because they
are nuclear modifiers of the RP, which might be within the potential focus
domain of the sentence. If the relative clause modifies a noun head that is
outside the potential focus domain – for example, an RP within an adverbial
clause in a clause juncture – the relative clause loses the possibility of being
part of the focus.
15.5 Conclusion
The study of RCs and clefts is enlightening in different ways and at

different levels. From a very abstract perspective, RC is a primary example
of the need for phrases in any grammatical theory. It is a structure – not
just a mere concatenation– that results from a meaningful combination of
symbols. The relative clause in RC is semantically and syntactically inde-
pendent from the matrix clause because it is embedded within a noun
phrase. The RP is an enclosure that fosters relations among the symbols
inside, preventing them from having certain relations with symbols out-
side. In addition, RC reveals the structural richness of human languages. As
defined here, relative clauses are embedded into noun phrases as periph-
eral modifiers of the nucleus. It turns out that different subordinations are
possible even in a single language, through a relative pronoun or no
subordinator at all. Furthermore, a typological perspective instructs us
about different structural conditions for the antecedent–pronoun co-
indexation relation. In the typologically more frequent externally headed
relative clause (Van Valin 2005; Comrie 1998), the antecedent noun is in
the matrix clause whereas the referentially dependent form or argument
slot belongs to the dependent clause. In internally headed relative clauses
the situation is the opposite, in that the full noun is part of the dependent
clause whereas the argument slot that takes its value from that noun is in
the matrix clause.
Clefts too illustrate very general properties of linguistic systems. The
perspective I have taken here to describe clefts bears on the question: how
deep does information structure carve into the grammatical system? In
other words, the issue is whether a system that is designed to satisfy
communicative needs operates at the grammatical core. A functionalist
linguist would predict that information structure goes all the way down
to the centre of the grammatical system, and this is precisely what it-clefts
show. The matrix clause does not carry the semantically rich content of the
sentence, which is in fact contributed by the embedded clause. This embed-
ded clause determines even the morphosyntactic realization of the attribu-
tive expression (typically a noun phrase but also a prepositional phrase) in
the matrix clause. This is what, in this chapter, has been called ‘inverted’
linking. However, the matrix clause conveys the focal element of the

612 LUIS PARÍS
assertion, that is, it provides the salient information structure content. In

short, information structure reverses the dependency relation that would be
predicted from mere semantic considerations.
The contrast between clefts and RCs highlights how two structures that
share striking similarities can be at the same time extremely different.
Both clefts and RCs contain an embedded clause that can be introduced by
a relative pronoun. Both of them hold a co-indexation relation between an
element in the matrix clause and another in the embedded clause. How-
ever, the relative clause in RC is a modifier of the antecedent noun,
whereas clefts involve an interclausal relation such that the cleft clause
is dependent on the matrix clause; it is in the periphery of the matrix core.
This analysis explains among other facts the different scope of quantifiers
in the matrix clause. Their different structural shape is systematically tied
to their different communicative function. The function of RCs revolves
around the identity of the participant expressed by the head noun. The
listener has to identify it with the help of the information carried by
the embedded clause, information that qualifies as a presupposition. The
function of clefts is centred on information structure. The matrix clause
conveys a participant whose identity, together with the role it plays in the
event introduced by the embedded clause, constitutes the focus of the
sentence. In this sense, the embedded clauses in RCs and clefts both carry
open propositions but the one in RC counts as a plain presupposition while
the one in clefts is an ‘operative presupposition’, which means that part of
it is in focus.
References
Bentley, Delia, Francesco Maria Ciconte and Silvio Cruschina. 2015. Existentials
and Locatives in Romance Dialects of Italy. Oxford: Oxford University Press.
Bird, Charles. 1968. Relative clauses in Bambara. Journal of West African
Languages 5: 35–47.
Chafe, Wallace. 1987. Cognitive constraints on information flow. In Robert
Tomlin (ed.), Coherence and Grounding in Discourse (Typological Studies in
Language 11), 21–52. Amsterdam: John Benjamins.
Cinque, Guglielmo. 2015. Three phenomena discriminating between ‘rais-
ing’ and ‘matching’ relative clauses. Semantics–Syntax Interface 2(1): 1–26.
Citko, Barbara. 2004. On headed, headless and light-headed relatives. Natural
Language and Linguistic Theory 22(1): 95–126.
Comrie, Bernard. 1998. Rethinking the typology of relative clauses. Language
Design 1: 59–86.
Davidse, Kristin. 2000. A constructional approach to clefts. Linguistics 38(6):
1101–1131.
Declerck, Renaat. 1988. Studies on Copular Sentences, Clefts and Pseudo-Clefts.
Berlin: De Gruyter Mouton.

Hedberg, Nancy. 2000. The referential status of clefts. Language 76(2):

891–920.
Huddleston, Rodney. 1984. Introduction to the Grammar of English. Cambridge:
Keenan, Edward L. 1976. Remarkable subjects in Malagasy. In Charles Li (ed.),
Subject and Topic, 247–301. New York: Academic Press.
Keenan, Edward L. and Bernard Comrie. 1977. Noun phrase accessibility and
universal grammar. Linguistic Inquiry 8: 63–99.
Kiss, Katalin. 1998. Identificational focus versus information focus. Language
74(2): 245–273.
and the Mental Representations of Discourse Referents. Cambridge: Cambridge
University Press.
Lambrecht, Knud. 2000. When subjects behave like objects: An analysis of
the merging of S and O in sentence focus constructions across languages.
Pavey, Emma L. 2004. The English It-Cleft Construction: A Role and Reference
Grammar Analysis. Unpublished dissertation, University of Sussex.
and syntax. In Ekatarina Raxilina and Jakov Testelec (eds.), Typology and the
Theory of Language: From Description to Explanation, 511–524. Moscow: Lan-
guages of Russian Culture.
Van Valin, Robert D., Jr. 2012. Some issues in the linking between syntax
and semantics in relative clauses. In Bernard Comrie and Zarina Estrada-
Fernández (eds.), Relative Clauses in Languages of the Americas: A Typological
Overview, Typological Studies in Language 102, 47–64. Amsterdam: John
Benjamins.
Van Valin, Robert and Randy LaPolla. 1997. Syntax: Structure, Meaning and
Function. Cambridge: Cambridge University Press.
Notes
1 Keenan and Comrie (1977) propose a universal hierarchy of RP functions

where the highest-ranking position represents the one most accessible to
license a relative clause.
(i) Subject > DO > IO > OBL > GEN(itive) > O(blique)COMP
The hierarchy should also be interpreted implicationally, namely, that

every language that accepts relative clauses for an RP down the hier-
archy, must also license RCs for higher-ranked positions in
the hierarchy.

614 LUIS PARÍS
2 I do not call ‘where’ an outright pronoun because its function in the RC is

adverbial, even if its antecedent might be unequivocally a noun (‘I know
the bar where you met Sally’).
3 I use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss
ing-rules.php) plus the following addition: ATV stands for active voice.
4 The only alternative semantic representation that I can think of would be
one in which the two LSs are coordinated and their shared argument co-
indexed.
(i) INGR quit′ (student1, program) ^ INGR fail′ (student1, test)
This representation is truth-conditionally equivalent to (10), but is less

influenced by the syntactic realization, and, crucially, it does not repre-
sent the attributive relation.
5 It might be appealing to attempt an analysis of the Bambara IHRC that
assigns a different semantics to it so that we can maintain the pattern of a
hierarchical relation between antecedent and co-indexed element. In
particular, an LS that obviates the attributive relation and simply relates
the meaning of the two clauses through coordination, as in (i).
(i) [see′ (ne, soi)] ^ [do′ (tye, Ø)] CAUSE [BECOME have′ (tye, xi)]
However, as was mentioned earlier, I contend that relative clauses are

semantically different from their pseudo-paraphrases in which coordin-
ation replaces the RC.
6 Realizing the pronoun with an accusative shape turns the sentence
ungrammatical (pret ¼ preterite).
(i) *Al que invitaste llegó tarde.

To.the that invite.pret arrive.pret late
‘The one to whom you invited came late.’
7 Citko (2004) labels the fact that this pronoun might have a case consistent
with one clause (the matrix) but not with the other (the embedded clause)
as a ‘non-matching’ property. She claims that ‘non-matching’ pronouns
are characteristic of ‘light-headed relative clauses’ and that sentences like
the one in (19) fall under this label. Her analysis of light-headed relatives
is couched in the Minimalist framework (she claims there is an empty
node being occupied by the raised pronoun). I prefer to maintain the
specific label ‘pronounless relative clauses’ in order to highlight the
differences between the two proposals.
8 Certain construction types might fall under the overarching notion of
clefts but they may instantiate a different information structure pattern.
In clefts, the embedded clause is not the focal element. However, pseudo-
cleft sentences involving a presentational clause of the type in (i) contain a
thetic or sentence-focus structure. The embedded clause is, thus, within
focus (Lambrecht 2000, Bentley et al. 2015).
(i) There was a man that came to my house.

9 The other possible interpretations of the copula are ‘predicative’, ‘iden-

tificational’ and ‘equative’ (Van Valin 2005: 48).
10 Certainly, ‘blame’ is ambiguous. The interpretation that is targeted here
is based on example (21), (26a) and it has the sense of a speech act verb,
an act that lasts an instant even if it has lasting consequences. Under this
interpretation, it cannot be an Accomplishment or an Activity (The teacher
blamed the bus driver *in/??for a month). Another interpretation of this verb
targets a mental state that can last for a possibly unlimited period of
time. This interpretation is especially suitable when used in the present
(John blames his parents for his academic failure).

16
Extraction Restrictions in
Complex Sentences
Mitsuaki Shimojo
Abbreviations

ASP aspect PrCS pre-core clot

DES desiderative PrDP pre-detached position
EVI evidential PoCS post-core slot
FP sentence-final particle PoDP post-detached position
IU information unit RP reference phrase
PFD potential focus domain
16.1 Introduction
For nearly five decades, extraction restrictions, also known as island

constraints, have been known to be an issue that every syntactic theory
should address, but they have remained elusive.* The elusiveness is
rooted in the intricate nature of the phenomena, which are not only
language- and construction-specific but also related to non-structural
grounds such as lexical-semantic and pragmatic factors, which influence
extractability. This chapter lays out the Role and Reference Grammar
(RRG) approach to extraction restrictions (Van Valin 1996; Van Valin and
LaPolla 1997) and shows how cross-linguistic and language-internal (i.e.
cross-constructional) variations are captured by the theory. In regard to
this purpose, this chapter also outlines a case study from Japanese, which
is radically different from English-type languages due to its relatively
restriction-free characteristics. Japanese exhibits restrictions in both
* I thank Delia Bentley and Robert D. Van Valin, Jr. for their valuable comments on earlier versions of the chapter.

Extraction Restrictions in Complex Sentences 617
in-situ and ex-situ constructions, and therefore, serves as a useful ground

for the RRG applications.
Since Ross’s (1967) discovery of the phenomena, syntactic constraints on
extraction, including those given in (1), have drawn much attention and
there has been further reformulation of the original proposal, beginning
with Chomsky (1973). In the course of the evolution of the Chomskyan
theory of syntax, island constraints have been subsumed under the general
principle of subjacency, which is stated in (2).
(1) a. Jon believes [NPthe rumour [Sthat Mary lost her phone]].
b. *What does Jon believe [NPthe rumour [Sthat Mary lost __ ]? (complex NP)
c. Jon talked to [NPthe neighbour [Swho bought the house recently]].
d. *What did Jon talk to [NPthe neighbour [Swho bought __ recently]]? (complex NP)
e. Jon ate a sandwich after [SMary made an omelette].
f. *What did [SJon eat a sandwich after [SMay made __ ]]? (adjunct)
g. [SThat Jon ate the sandwich] was obvious.
h. *What was [S[Sthat Jon ate __ ] obvious]? (subject)
(2) Subjacency condition (Chomsky 1973)

Extraction cannot cross more than two bounding nodes (NP, S) in a
single movement.
In this formulation, it has been assumed that these constraints are purely
structural with no reference to a semantic, pragmatic, or other basis. There-
fore, they have been taken as evidence of autonomous syntax and ultimately
of a theory of universal grammar.
While there has been much theory-internal debate with respect to the
status of the subjacency condition (see Yoshimura 1992 and references cited
therein), there have also been proposals to account for the constraints,
external to the syntactocentric theories, such as semantic and pragmatic
approaches (Erteschik-Shir 1973; Erteschik-Shir and Lappin 1979; Haig 1996;
Kuno 1987; Kuno and Takami 1993; Shimojo 2002; Van Valin 1996; Van Valin
and LaPolla 1997), cognitive and performance-based proposals (Deane 1991,
1992; Hofmeister 2007; Kluender 1990; Kluender and Kutas 1993), and pars-
ing and expectation-based claims (Chaves 2013). In short, the subjacency
accounts are problematic because the constraints are sensitive to a range of
factors beyond pure syntactic grounds. Furthermore, for a number of theor-
ies, the subjacency account is problematic since it assumes movement,
which is a theory-internal assumption, and therefore incompatible with
monostratal and non-transformational theories such as RRG.
In addition, there is difficulty with a movement-based approach in
accounting for constraints which do not involve movement without added
theory-internal stipulations. Lakhota is a language that does not favour the
subjacency account. Van Valin (1996: 36–37) points out that Lakhota blocks
wh-questions formed within a complex NP despite its wh-in-situ characteris-
tics. A sentence containing a word such as táku or tuwá and the question

618 MITSUAKI SHIMOJO
marker he in Lakhota is ambiguous between wh-question and yes/no ques-

tion interpretations because the words are ambiguous between wh-word and
indefinite-specific pronoun readings (3a). However, the ambiguity is resolved
with complex reference phrases (RPs) and adverbial clauses. As shown in
(3b), the wh-question reading is unacceptable, and táku in the adverbial
clause can be interpreted only as ‘something’; thus, this is analogous to
the subjacency constraint observed in English.
(3) a. [Tuwá thaló ki manú] iyúkčą he?

who/someone meat the steal think q
‘Who does he think stole the meat?’ or ‘Does he think someone stole the meat?’
b. [Wičháša ki táku yúte] ečhúhą, tha-wíču
man the *what/something eat while his-wife
ki mní ikíčíču he?
the water get.for q
‘While the man was eating something, did his wife get him water?’
*‘What did his wife get him water, while the man was eating?’
(Intended wh-question: ‘Did his wife get him water, while the man was
eating what?’)
(Van Valin 2005: 277)
In order to account for the restriction in (3b), the movement-based account
requires a movement at abstract levels so that subjacency applies to lan-
guages such as Lakhota (Chomsky 1986); however, such an account lacks a
theory-external appeal. Furthermore, since Lakhota-type languages, which
have wh-in situ, exhibit extraction constraints, movement or filler-gap
dependency is not an essential feature to account for the restrictions. For
this reason, Lakhota-type languages also pose a challenge to any approach
which claims that processing difficulties or parsing preferences are rooted
in filler-gap dependency as an underlying principle.
Lastly, there are languages which do not obey subjacency yet exhibit
construction-specific restrictions and/or constraints under certain semantic
and pragmatic conditions. These languages include Chinese (Jin 2013),
Danish (Erteschik-Shir and Lappin 1979; Jensen 2001), Korean (Hong 2003),
Swedish (Allwood 1982) and Japanese (Haig 1996; Shimojo 2002). Japanese
will be discussed at length later.
16.2 Role and Reference Grammar Accounts
RRG combines structural and pragmatic grounds to capture both language-

internal and cross-linguistic variations in extraction restrictions.1 The rele-
vant pragmatic property is represented in the focus structure projection of a
sentence, which indicates the domain of the sentence that represents or
may represent the focus, that is, pragmatic assertion in the sense of Lam-
brecht (1994: 52). There are cross-linguistic variations as to how rigid or
flexible a language is with respect to the placement of focus (Van Valin
1999). In Italian, for example, a preverbal non-contrastive focal RP is not
allowed, and therefore, a focal subject RP needs to be postverbal (i.e. word

ordering is relatively flexible). English exhibits the opposite pattern with a

relatively rigid SVO ordering and flexible placement of focus within the
sentence; either the subject or the object can be focal in the same word
order (see Chapter 11 in this volume).
With respect to focus structure for complex sentences, the cross-linguistic
concern is to what extent the focus domain is extended into subordinate
clauses. Some languages such as Polish restrict the focus domain to matrix
clauses only, and in other languages such as English, the focus domain
includes certain types of subordinate clauses (Van Valin 1996). In RRG,
language-internal variation in terms of the focus domain is described struc-
turally as stated in (4).
(4) The potential focus domain in complex sentences

The potential focus domain extends into a subordinate clause if and only if the
subordinate clause is a direct daughter of (a direct daughter of ) the clause node
which is modified by the illocutionary force operator (Van Valin 2005: 275).
For example, an object complement clause is a direct daughter of a clause

node; therefore, the potential focus domain (PFD), where a focal element
may appear, extends into the subordinate clause. In contrast, this is not the
case with an adverbial subordinate clause, which is in the periphery to the
core, and only the subordinate clause as a whole, which is a constituent of
the matrix clause, may function as an information unit. Figures 16.1 and
16.2 show the contrast between the two types of subordination with respect
to the scope of the PFD, which is represented by the dotted line.
SENTENCE
CLAUSE
CORE
RP NUC RP
PRED CLM CLAUSE
V CORE PERIPHERY
Kim told Pat that RP NUC PP
PRED ADV
she will arrive at the party late.
IU IU IU [IU IU IU IU]
IU
SPEECH ACT
Figure 16.1 Potential focus domain in clausal (daughter) subordination (Van Valin 2005: 214)

SENTENCE
CLAUSE
CORE PERIPHERY
RP NUC RP PP
PRED NUC CLAUSE
V P CORE
Pat saw Kim RP NUC PP
PRED
after she arrived at the party
IU IU IU IU
SPEECH ACT
Figure 16.2 Potential focus domain in ad-core subordination (Van Valin 2005: 216)
In both types of subordinate clause shown in the figures, the subordinate

clause as a whole can be an information unit (IU), which can be focal, as in
the case of an answer to ‘what did Kim tell Pat?’ and ‘When did Pat see
Kim?’. However, the ad-core subordinate clause in Figure 16.2 does not allow
internal constituents to be distinct information units because they cannot
be the focus individually. This is shown in the question–answer pairs in (5).
The unacceptable answer in (c) shows that the internal constituents of the
subordinate clause are outside of the PFD.
(5) Q: Did Pat see Kim after shei arrived at the party?
A: a. No, Sally.
b. No, before.
c. *No, shei left.2
The cross-linguistic and language-internal observations in terms of

focus domains correlate with extraction restrictions. First, languages like
Polish, which restrict the focus domain to matrix clauses, prohibit extrac-
tion out of tensed subordinate clauses; therefore, extraction is a matrix
phenomenon in these languages. On the other hand, English-type lan-
guages allow extraction out of a subordinate clause within the PFD. Thus,
extraction is possible out of an object complement clause (e.g. where did
John tell Mary that he will arrive __?), but not out of an adverbial subordin-
ate clause (e.g. *where did John see Mary after he arrived __?). In other words,
an extraction site must be within the PFD and this is captured by the
principle stated in (6).

(6) General restriction on extraction constructions (Van Valin 1996: 54)

The displaced element (or the in situ question word in a language like Lakhota)
must be linked to an argument position in the semantic representation of a
clause within the PFD of the illocutionary force (IF) operator.
From a cross-linguistic perspective, extractability can be captured in terms

of the scale given in (7). Languages like Polish allow extraction only from a
matrix clause and this is the most inflexible type. Languages like English
and Lakhota are more flexible, allowing extraction from certain types of
subordinate clause. The scale would be complete with another type which
allows extraction out of a greater range of subordinate clauses. Japanese is of
this type and the following section discusses how the RRG principle given in
(6) applies to Japanese.
(7) Range of extractability: cross-linguistic variations

Matrix clause only Polish . . .
Subordinate clauses: direct daughter of clause only English, Lakhota . . .
More subordinate clauses Danish, Japanese, Swedish . . .
16.3 Application of the RRG Analysis to Japanese
The cross-linguistic variations outlined thus far raise a question about the
validity of the RRG principles in a language in which extraction is
allowed more freely, and therefore, this section discusses extraction
restrictions in Japanese. Studies on extraction constraints have been
centred on English-type languages, in which extraction constraints are
more or less structurally predictable. In relatively island-free languages,
restrictions are typically described on a semantic and pragmatic basis,
and Japanese is no exception (see the references cited for these languages
in Section 16.1).
16.3.1 Possible Extraction in Complex Sentences

The possible range of extraction in Japanese is exemplified by the following
three major extraction constructions: wh-question, relativization and
topicalization.
(8) Wh-question formation in a relative clause

seefu-wa [dare-ga sekkee-shita] doroon-o koohyoo-shimashita ka
government-top who-nom design-do.pst drone-acc release-do.pst q
‘The government announced the drone who designed?’
(9) Relativization out of a relative clause

[seefu-ga [ __ sekkee-shita] doroon-o koohyoo-shita] enjinia
government-nom design-do.pst drone-acc release-do.pst engineer
‘The engineer who the government announced (s/he) designed the drone.’

(10) Topicalization out of a relative clause

sono enjinia-wa [seefu-ga [ __ sekkee-shita]
that engineer-top government-nom design-do.pst
doroon-o koohyoo-shita
drone-acc release-do.pst
‘The engineer, the government announced (s/he) designed the drone.’
Furthermore, the extensive range of possible extraction is exemplified by

the following wh-questions formed in different types of subordinate clause: a
noun complement, a sentential subject and an adjunct clause (relativization
and topicalization out of these subordinate types are also possible; see
Shimojo 2002).
(11) Wh-question formation in a noun complement

jon-wa [dare-ga doroon-o sekkee-shita] toiu joohoo-o
Jon-top who-nom drone-acc design-do.pst comp information-acc
utagatte-imasu ka
doubt-asp q
‘Jon doubts the information that who designed the drone?’
(12) Wh-question formation in a sentential subject

[dare-ga doroon-o sekkee-shita koto]-ga yosoogai deshita ka
who-nom drone-acc design-do.pst nmlz-nom unexpected cop.pst q
‘It was unexpected that who designed the drone?’
(13) Wh-question formation in an adjunct clause

[paatii-ni dare-ga kita toki] jon-ga deteikimashita ka
party-dat who-nom come.pst when Jon-nom leave.pst q
‘Jon left when who came to the party?’
16.3.2 Focus Domain in Complex Sentences

The focus-based RRG principle discussed in Section 16.2 predicts the associ-
ation between the focus domain and a possible extraction site. This predic-
tion is borne out in Japanese because the potential focus domain covers the
individual constituents of a subordinate clause regardless of the principle
given in (4). This is demonstrated by the following tests.
Only the asserted part of an utterance can be interpreted as being negated.
If the constituent can be negated in a conversational exchange, then it is a
possible focus (i.e. in the potential focus domain). The example in (14) shows
the negation test applied to a relative clause. The possible negation of the
information expressed by the relative clause shows that the information
may be interpreted as part of the assertion.
(14) A: seefu-ga [doroon-o sekkee-shita] enjinia-o koohyoo-shita

government-nom drone-acc design-do.pst engineer-acc release-do.pst
‘The government announced the engineer who designed the drone.’
B: iya, misairu da
no missile cop
‘No, a missile.’

As Van Valin and LaPolla (1997: 626) point out, in English, an adjunct wh-
question is ambiguous in a complex sentence in which the PFD extends over
the subordinate clause, as shown in (15a), because ‘when’ can be interpreted
as modifying either the matrix clause or the object complement. However,
ambiguity does not arise in a subordinate clause which is outside a PFD, as
in (15b).
(15) a. Wheni/j did Skinner say __i that Krycek would be at the missile silo __j?
b. Wheni/*j did Skully interview the witness __i who saw the alien spacecraft
in the silo __j?
In contrast, the corresponding Japanese sentences are both ambiguous, as

shown in (16), which indicates that both the object complement and the
relative clause may be part of the assertion (i.e. within a PFD) in Japanese.3
(16) a. itsu kuroichekku-ga kakunooko-ni iru-to sukinaa-ga itta?

when Krycek-nom silo-loc exist-quot Skinner-nom say.pst
‘Wheni/j did Skinner say __i that Krycek would be at the missile silo __j?’
b. itsu kakunooko-de uchuusen-o mita mokugekisha-to
when silo-loc spacecraft-acc see.pst witness-com
sukarii-ga menkai-shita?
Skully-nom interview-do.pst
‘Wheni/j did Skully interview the witness __i who saw the alien spacecraft
in the silo __j?’
On the one hand, the PFD excludes the pre-detached position (PrDP) and
the post-detached position (PoDP), which contain a sentence-initial topic and
a sentence-final (i.e. post-predicative) topic respectively. On the other hand,
the focus domain includes the pre-core slot (PrCS), which contains a
sentence-initial narrow-focus (i.e. a narrow-focus subject or a pre-posed
narrow-focus), and the post-core slot (PoCS), which contains a post-
predicative focus. Figure 16.3 shows the layered structure of the clause with
respect to the PFD.
Japanese uses post-nominal markings that correlate with the information
structuring of the sentence. The topic of a sentence, if any, is outside the PFD
SENTENCE
PrDP CLAUSE PoDP
PrCS CORE PoCS
RP RP NUC
PRED
SPEECH ACT
Figure 16.3 Layered structure of the clause and potential focus domain

and it may be overtly marked with the topic marker wa. In contrast, if a
nominative argument is marked with the nominative marker, it is within
the PFD, whether it is narrow focus (PrCS) or part of a broad-focus (RP). The
direct evidence for a detached position being outside of the PFD comes
from the observation that a wh-question word cannot be topicalized, as
shown in (17).4
(17) dare-ga/*wa sono uchuusen-o mimashita ka

who-nom/*top that spacecraft-acc see.pst q
‘Who saw the spacecraft?’
Also, as shown in (18), it is slightly more awkward to negate information

expressed by an element which is part of the topic than to negate infor-
mation expressed by a non-topic (cf. (14)).5
(18) A: [doroon-o sekkee-shita] sono enjinia-wa seefu-ni

drone-acc design-do.pst that engineer-top government-dat
koohyoo-sare-ta
release-do.pass-pst
‘The engineer who designed the drone was publicized by the government.’
B: ?iya, misairu da
no missile cop
‘No, a missile.’
The functional principle given in (6) predicts that an extraction site is not
permitted within a topic phrase, which is outside the PFD, and this is indeed
the case, as shown in (19). The sentence with the topicalized complex RP is
not felicitous when no particular presupposition for the sentence is present
(for example, when the sentence is given out of the blue).
(19) [dare-ga shuppan-shita] hon-ga/*wa yoku ureru?

who-nom publication-do.pst book-nom /*top well sell
‘Do books that who published sell well?’
This is also the case with relativization and topicalization out of a topica-
lized complex RP, as shown in (20) and (21) respectively.
(20) [[ __ shuppan-shita] hon-ga/*wa yoku ureru] joyuu

publication-do.pst book-nom/*top well sell actress
‘The actress who books that (she) publishes sell well.’
(21) sono joyuu-wa [ __ shuppan-shita] hon-ga/*wa yoku ureru

that actress-top publication-do.pst book-nom/*top well sell
‘The actress, books that (she) publishes sell well.’
However, the topicalization of an extraction site is possible only when the

sentence is intended to be overtly contrastive. In other words, the topic
marking in the examples above would be possible if the question in (19)
could be interpreted as contrastive with respect to books which someone

publishes as opposed to books which others publish, and if those in (20) and
(21) can be interpreted as contrastive with respect to books as opposed to
other things. In fact, the imposed contrastiveness associated with an extrac-
tion site further supports the principle in (6).
Erteschik-Shir (2007) claims that contrast is contextually constrained to
occur only if a contrast set is available, and because contrast is represented
by singling out a subset of a whole to separate it from the remaining
subset, the singled-out subset represents a subordinate focus, which is
embedded in the matrix topic. An example is given in (22). ‘John’ is part
of the presupposition for (22A); however, it is a newly singled-out member
of the previously given set (hence it is focal). In the formal notation, a
contrastive topic represents the subordinate focus structure, as shown on
the right-hand side. In other words, a (matrix) topic which contains a focus
simultaneously represents a contrast, and this corroborates the principle
in (6) because a topic must be contrastive (i.e. focal) in order to contain an
extraction site.
(22) Q: Tell me about your brothers John and Bill.

A: JOHN is the smart one. Contrastive topic: [{Johnfoc, Bill}top ] top
A topic which allows an extraction site within cannot be in the detached

position, which is outside the PFD; it must be within the PFD of the sentence
as it contains a focus element, being placed within the clause, either core-
externally in the pre-core (or post-core) slot or core-internally.6
16.3.3 Extraction Restrictions in Complex Sentences

16.3.3.1 Relativization and Topicalization
The possible range of extraction in Japanese has been outlined thus far, and
now discussions of extraction restrictions are in order. Despite the high
degree of extractability, the language is not free from restrictions (Kuno
1973; Haig 1996; Shimojo 2002). Examples of unacceptable topicalization
and relativization are given in (23) and (24) respectively.
(23) *sono kuruma-wa keesatsu-ga [__ nusunda] otoko-o taiho-shita

that car-top police-nom steal.pst man-acc arrest-do.pst
‘The car, the police arrested the man who stole (it).’
(24) *[keesatsu-ga [__ nusunda] otoko-o taiho-shita] kuruma

police-nom steal.pst man-acc arrest-do.pst car
‘a car which the police arrested the man who stole (it).’
These cases of unacceptable extraction are not predictable by the focus-

based principle in (6) since relative clauses in Japanese are within a PFD, as
discussed earlier. The observed unacceptability is due to the aboutness
condition that is not met in these cases. The condition stated in RRG terms
is given in (25).

(25) Pragmatic-aboutness condition on topicalization and relativization

The sentence fragment following a topical element in the pre-core slot or a
restrictive relative clause must be pragmatically interpretable as being about
the pre-core slot element or the head noun.
(Van Valin and LaPolla 1997: 627)
The constraints are observed when it is difficult to obtain the interpretation

that the sentence or the relative clause is about the displaced entity. In other
words, when a subordinate element is relativized or topicalized, the ‘inter-
vening’ matrix clause as well as the subordinate clause (i.e. the extraction
site) must be relevant to the displaced topic or the relative clause head. This
general principle, which relates to the notion of relevance, is valid for
extraction from an RP in English (Kuno 1987).
(26) a. Who did John write a book about?

b. ?Who did John destroy a book about?
c. ?Who did John lose a book about?
The examples in (26b, c) show degraded extractability. In these cases, it is

difficult to obtain the interpretation that the sentence is about the displaced
element. As Chaves (2013: 308) puts it, while the action of writing is imme-
diately relevant to the book’s topic, destroying or losing the book is not. In
other words, in these questionable cases, the displaced element is not
relevant to the assertion made by the matrix clause. In RRG, this aboutness
principle relates to the principle of focus domain given in (6) because the
verbs in (26b, c) are informationally distinctive and therefore naturally draw
the focus, preventing the object RP from being the actual focus (Van Valin
2005: 288–289).
The aboutness condition is further demonstrated by the contrast between
the two Japanese examples in (27).
(27) a. *[keesatsu-ga [ __ nusunda] otoko-o taiho-shita] kuruma (cf. (23))

‘a car which the police arrested the man who stole (it)’
b. [[ __ kite-iru] yoohuku-ga yogorete-iru] shinshi
wear-asp clothes-nom get.dirty-asp gentleman
‘a gentleman who the clothes that (he) is wearing are dirty’ (Kuno 1973: 239)
(27a) is unacceptable because the stolen car is not immediately relevant to

the action of arresting per se (the ‘car’ in this case is more naturally
interpreted as a police car used for the arrest, the vehicle that is directly
relevant to the action of arresting). In (27b), on the other hand, the state of
being dirty is immediately relevant to the person who is wearing the dirty
clothes. Hence, this relativization is perfectly acceptable.7
Because Japanese allows the so-called gapless relative clauses, in which the
head noun is relatable to the relative clause only semantically or pragmatic-
ally (Matsumoto 1997), the truncation diagnostic given in (28) serves as a
useful means to see how relatable a head noun is to the matrix clause.

(28) The truncation test for aboutness (Shimojo 2002: 77)

Relativization and topicalization out of embedded clauses are possible if the
relative or topic construction is acceptable even without the embedded clause
in which the extracted NP functions.
In essence, if the displaced RP is relevant to the intervening matrix clause,

the intended relevance should be inferable or at least should not be contra-
dictory even if the clause containing the extraction site is omitted. The
truncated versions of (27a, b) are given in (29a, b) respectively.
(29) a. #[keesatsu-ga otoko-o taiho-shita] kuruma

police-nom man-acc arrest-do.pst car
‘a car which the police arrested the man (who stole (it))’
b. [yoohuku-ga yogorete-iru] shinshi
clothes-nom get.dirty-asp gentleman
‘a gentleman who the clothes (that (he) is wearing) are dirty’
The possible interpretation of (29a), the truncated version, is ‘a car which

the police used in arresting the man’, and this reading contrasts with the
intended reading, which is shown by the translation. On the other hand,
the possible reading of (29b) is consistent with the intended reading, and the
observed contrast between the two truncated versions corresponds with the
contrast in acceptability between the two original versions in (27).
Inoue (1976) claimed a subject bias in extraction, and Hasegawa (1985: 292)
proposed as a structural principle that the relativization or topicalization out
of a relative clause is allowed if the displaced phrase is the subject of the
(lower) relativized clause and the head of the relative clause is the subject of
the higher clause. Although it has since been shown that this is problematic
as a principle (see Shimojo 2002: 74–5 and the references cited therein), the
said subject bias, if taken as a tendency, is not so inconsistent with what
the pragmatic-aboutness condition predicts. Assuming a subject represents
the target of predication, the displaced element functions as (part of ) the
target of predication if the displaced element is the subject of the lower
relative clause which is in turn the subject of the higher clause. The example
of an acceptable extraction given in (27b) represents this structure.
However, it should be noted that subjecthood does not warrant required
aboutness because aboutness is also affected by lexical choice. This point is
demonstrated by the relativization shown in (30). While this example repre-
sents the optimal structure per Hasegawa (1985), the verb choice for the
inner relative clause affects the extractability.
(30) [[ __ kaita/*yonda] hon-ga eegaka-sare-ta] jon

write.pst/read.pst book-nom cinematization-do.pass-pst Jon
‘Jon who a book which (he) wrote/*read was made into a movie’
Truncation of this example results in [hon-ga eegakasareta] jon ‘Jon who a book
was made into a movie’. Possible interpretations of this truncation are Jon as
the author, editor, translator, etc. of the book (as translated as ‘Jon whose

book was made into a movie’), not a mere reader, and this contradicts the
intended reading of [[ __ *yonda] hon-ga eegakasareta] jon.
16.3.3.2 Postposing
Sentence elements are also displaced through the so-called postposing
construction. While Japanese is considered a rigid verb-final language, it
allows sentence elements to be in post-predicative positions, typically in
spoken Japanese (Shimojo 1995).8 An example is shown in (31b), in which
the nominative argument appears in the postverbal position, in contrast
with (31a), which represents the canonical ordering of arguments (the post-
posed elements are underlined in the examples).
(31) a. keesatsu-ga otoko-o taiho-shita yo

police-nom man-acc arrest-do.pst fp
b. __ otoko-o taiho-shita yo keesatsu-ga
man-acc arrest-do.pst fp police-nom
‘The police arrested a man.’
As shown in Figure 16.3, there are two positions for post-predicative elements
in Japanese, PoCS and PoDP. While both are structurally marked positions in
the verb-final language, they represent distinct focus-structure properties. The
PoDP is a post-predicative topic position, the post-predicative counterpart of
PrDP, and for this reason, the preceding discussion of topicalization applies
here. For the PoDP-type (topic) postposing, the preceding part of the sentence
including the matrix clause must be about the displaced element. The
examples in (32) show the correspondence between acceptable relativization
(a), topicalization (b) and PoDP-type postposing (c).
(32) a. [[ __ kite-iru] yoohuku-ga yogorete-iru] shinshi

wear-asp clothes-nom get.dirty-asp gentleman
‘a gentleman who the clothes that (he) is wearing are dirty’
b. ano shinshi-wa [ __ kite-iru] yoohuku-ga yogorete-iru
that gentleman-top wear-asp clothes-nom get.dirty-asp
‘That gentleman, the clothes that (he) is wearing are dirty.’
c. [ __ kite-iru] yoohuku-ga yogorete-iru yo ano shinshi-wa
wear-asp clothes-nom get.dirty-asp fp that gentleman-top
‘That gentleman, the clothes that (he) is wearing are dirty.’
On the other hand, because a postverbal element in PoCS does not repre-
sent a topic, the aboutness condition is not a requirement for the PoCS-type
(focus) postposing.9 For example, the postposing in (33b) is acceptable des-
pite the unacceptability of the corresponding relativization.
(33) a. *[keesatsu-ga [ __ nusunda] otoko-o taiho-shita] kuruma (¼27a)

‘a car which the police arrested the man who stole (it)’
b. keesatsu-ga [ __ nusunda] otoko-o taiho-shita yo kuruma-o
police-nom steal.pst man-acc arrest-do.pst fp car-acc
‘The police arrested the man who stole a car.’

However, it has been observed that this type of postposing is sensitive to the
focus structure of a sentence. Simon (1989) gives the examples in (34) to
point out that transparency defined by genericness correlates with the
gradient acceptability. The postposing is acceptable with a matrix copula
(34a), and it is least acceptable with a specific action verb (34c).
(34) a. kore-wa [__kinoo mottekita] wain da yo mari-ga

this-top yesterday bring.pst wine cop fp Mari-nom
‘This is wine which Mari brought yesterday.’
b. ?* [__ kinoo mottekita] wain-ga aru yo mari-ga
yesterday bring.pst wine-nom exist fp Mari-nom
‘There is wine which Mari brought yesterday.’
c. *[__ kinoo mottekita] wain-o nomu yo mari-ga
yesterday bring.pst wine-acc drink fp Mari-nom
‘I’ll drink the wine which Mari brought yesterday.’
(Simon 1989: 166)
Also, the acceptability of postposing is affected by the particular focus

placement determined by the preceding context. The postposing in (35B) is
acceptable in isolation or in a context such as (35A). The relative clause in
(35B) represents the actual focus and the postposed argument functions
within the actual focus domain.
(35) A: jon-wa dare-ni aitagatte-ru no?

John-top who-dat meet.des-asp fp
‘Who does John want to meet?’
B: [__i __j kaita] hitoi-ni aitagatte-ru n da yo ano hon-oj
write.pst person-dat meet.des-asp nmlz cop fp that book-acc
‘(He) wants to meet the person who WROTE THE BOOK.’
However, in the context given in (36A), the relative clause is outside the
actual focus domain, and the matrix (bridging) clause represents the narrow
focus, which clearly impairs its postposability.10
(36) A: John-wa [ano hon-o kaita] hito-ni atta no?

John-top that book-acc write.pst person-dat meet.pst fp
‘Did John meet the person who wrote the book?’
B: *iya, [__i __j kaita] hitoi-ni aitagatte-ru n da yo ano hon-oj
no write.pst person-dat meet.des-asp nmlz cop fp that book-acc
‘No, (he) WANTS TO MEET the person who wrote the book.’
The preceding observations point to the functional requirement in PoCS-

type postposing that the displaced entity in the PoCS must function in the
part of the utterance which represents the assertion.11 This requirement
predicts that if the matrix clause represents a narrow focus, as in (36B), the
postposing is unacceptable. This basic requirement for displacement of non-
topic entities also applies to wh-questions, which are discussed in the
following section. Overall, postposing exhibits ambivalent properties. On
one hand, PoDP-type postposing is analogous to relativization and topicali-
zation because the displaced entity functions as a topic of the entire

sentence. The PoCS-type postposing, on the other hand, is similar to wh-

questions because the displaced element (or the in-situ question word) must
be associated with the focus of the sentence. In this regard, Takita’s (2014:
139) example given in (37) is worth a brief discussion. Takita states that the
postposing is unacceptable if the postposed argument ‘that book’ maintains
the accusative marking, but the sentence becomes acceptable if the case
marker is absent (i.e. if it is a ‘bare-topic’).12
(37) [taroo-ga __ suteta kara] hanako-ga totemo okotte-iru yo

Taro-nom discard.pst because Hanako-nom very get.angry-asp fp
ano hon-{*o/Ø}
that book-acc/zero.particle
‘Because Taro discarded that book, Hanako is very angry.’
In RRG terms, the observation above is captured as follows. The postpos-

ing with the zero marking for the postverbal argument is the PoDP type
because the postverbal argument is readily interpreted as a topic. The
postverbal RP is referential, and the lack of case marking is characteristic
of a topic. Also, the ‘book’ can be taken as the entity which the whole
sentence is relevant to, that is, that Hanako is very angry about the book
which was discarded. Yet, the overt accusative marking of the postverbal
argument de-topicalizes the argument and therefore shifts it to the PoCS;
hence, the postposability is degraded by the intervening matrix clause,
which represents assertion.
16.3.3.3 Wh-questions
While wh-questions in complex sentences in Japanese are generally accept-
able, some restrictions have been observed. Consider the examples in (38),
taken from Haig (1979: 90–91, grammaticality judgement original).
(38) a. ?* [dare-ga hanako-ni kureta] inu-ga shinde-shimatta?

who-nom Hanako-dat give.pst dog-nom die-asp.pst
‘The dog that who gave to Hanako died?’
b. ?* kono ko-wa [ __ hanako-ni __ kureta] inu-ga shinde-shimatta
this child-top Hanako-dat give.pst dog-nom die-asp.pst
‘This child, the dog which (he) gave Hanako died.’
The wh-question formed out of the relative clause in (38a) is unacceptable.

Given the unacceptability of the corresponding topicalization in (38b),
Haig claims that the questioned noun phrases raise an issue that essen-
tially relates restrictions on wh-questions to the aboutness condition dis-
cussed earlier. Kuno (1987: 23) has proposed the Topichood Condition for
Extraction, which subsumes wh-questions in English such as (26b, c). How-
ever, restrictions observed in Japanese wh-questions are of a different
nature from those in relativization and topicalization, which are condi-
tioned on aboutness. Consider the examples in (39), taken from Hasegawa
(1981: 281).

(39) a. *sono yoohuku-wa [ __ kite-ita] shinshi-ga yukuehumee da

that clothes-top wear-asp.pst gentleman-nom missing cop
‘The clothes, the gentleman who was wearing (them) is missing.’
b. [ __ nani-o kite-ita] shinshi-ga yukuehumee-na no?
what-acc wear-asp.pst gentleman-nom missing-cop fp
‘The gentleman who was wearing what is missing?’
The unacceptable extraction in (39a), and unacceptable truncation yoohuku-

wa shinshi-ga yukuehumeeda ‘the clothes, the gentleman (who was wearing
(them)) is missing’, shows that the topicalized element ‘clothes’ is not
immediately relevant to the gentleman’s being missing. Despite this, the
corresponding wh-question is acceptable (39b), and this undermines the
claim which relates wh-words to aboutness. In this regard, Van Valin and
LaPolla (1997: 627–629) argue that the function of a displaced element in a
wh-question (i.e. the wh-word) is focus, and therefore, they must function in
the PFD. This claim is valid for wh-questions in Japanese despite the wh-in situ
characteristics. What influences extractability in Japanese wh-questions are
lexical-semantic factors which interact with the PFD. As discussed thus far,
the default range of potential focus domain is language-specific. The struc-
tural constraint given earlier in (5) represents the default range of PFD in
languages like English, while in Japanese, the PFD excludes only detached
positions. Yet, in both languages, the PFD interacts with lexical-semantic
factors, as Van Valin and LaPolla (1997: 630) state:
lexical semantic factors may also influence the potential focus domain,
both in terms of preventing a position in the potential focus domain
from being the actual focus domain and of overriding the principle [in
(5)] and permitting the actual focus domain to be in structural
configurations where it would otherwise be impossible.
An object complement is part of the PFD by default since it is a direct

daughter of the clause node. However, it has long been observed, as shown
in (40a), that extraction is blocked by so-called non-bridge verbs, which
denote a particular manner, such as ‘murmur’, ‘whisper’ and ‘shout’, unlike
bridge verbs such as ‘say’, ‘think’ and ‘believe’ (Erteschik-Shir 1973; Müller
1995; Takami 1992). In RRG terms, these semantically highlighted verbs shift
the focus and reduce the PFD away from the complement clause, hence they
block the wh-question (Van Valin 1996: 50). The same argument applies to
the unacceptable extraction when a bridge verb is used with a manner
adverb (Kennedy 1989), as shown in (40b). The manner adverb causes the
same focus shifting, preventing the object complement from being the
actual focus domain.
(40) a. What did Jon say/*murmur/*whisper that Mary had bought?

b. *What did Jon say abruptly that Mary stole?
Analogous lexical-semantic factors are responsible for restrictions on wh-

questions in Japanese, as seen in (38a). However, wh-question formation is

affected only in a certain type of relative clause. First, consider Hasegawa’s

(1989) proposal cited in (41), with respect to different readings of a wh-
question formed in a relative clause.
(41) Two possible readings of a wh-question formed in a relative clause
(i) Narrow reading: what is questioned (i.e. the wh-phrase) is what is to be
identified by the question.
(ii) Broad reading: what is questioned is NOT what is to be identified by the
question. What is to be identified is represented by the relative clause
head noun.
According to Hasegawa, the interpretation depends on two separate factors:

the referability of the head noun and the inherent topic-worthiness of the
wh-expression. The former bears immediate relevance on the present discus-
sion. Now consider the contrast between (42) and (43).
(42) Q: [dare-ga kaita] shoosetsu-ga yoku uremasu ka
who-nom write.pst novel-nom well sell q
‘Novels that who writes sell well?’
A: murakami-haruki desu
Murakami.Haruki cop
‘(It’s) Haruki Murakami.’
A′: [murakami-haruki-ga kaita] shoosetsu desu
Murakami.Haruki-nom write.pst novel cop
‘(It’s) the novel which Haruki Murakami wrote.’
(43) Q: [dono kyooju-ga suisenshite-iru] hito-ga saiyoo-sare-soo desu ka

which prof.-nom recommend-asp person-nom employment-do.pass-evi cop q
‘A person that which professor has recommended is likely to be hired?’
A: *suzuki-kyooju desu
Suzuki-prof. cop
‘(It’s) Prof. Suzuki.’
A′: [suzuki-kyooju-ga suisenshite-iru] hito desu
Suzuki-prof.-nom recommend-asp person cop
‘(It’s) the person that Prof. Suzuki has recommended.’
(Nishigauchi 1986: 74)
Wh-questions as in (42) have a narrow reading, because the head of the
relative clause shoosetsu ‘novel’ is non-referential and, therefore, what is to
be identified by the question is who it is that sells their novels, rather than
the particular novels which the writer publishes. On the other hand, wh-
questions as in (43) have a broad reading, because the head noun hito
‘person’ is referential, that is, what is to be identified by the question is
who the candidate is, and this identification of the person is done by way of
identifying the recommender of the candidate. Hence, Hasegawa argues
that the short answer, which supplies just the value for the wh-word, is
awkward in the context of (43). As expected, for wh-questions such as (43) to
be felicitous, there must be a shared assumption in the context that there is
a set of candidates and each of them is recommended by a professor. The
purpose of the wh-question is to identify a particular candidate via

identifying a particular candidate–recommender pair. Wh-questions as in

(43) are thus virtually synonymous with ‘which’-questions such as ‘which
candidate is most likely to get the position?’
The distinction between narrow reading and broad reading sheds light on
why constraints are observed in some wh-questions. A narrow reading con-
tains a single focus of question, since what is questioned (i.e. the wh-phrase)
is what is to be identified by the question. On the other hand, a broad
reading represents two foci because what is to be identified by the question
(i.e. the relative clause head noun) represents a focus of the question, and
this focus element is additional to the canonical focus of the question (i.e.
the wh-phrase). In other words, wh-questions which represent a broad read-
ing are functionally complex, and it is this type of wh-question that is subject
to restrictions due to lexical-semantic factors.
Haig’s example given earlier in (38a) is a broad-reading question with the
referential head noun ‘dog’, which represents what is to be identified by
the question; thus, for this question, [dare-ga kureta] inu ‘the dog that who
gave’ is virtually synonymous with ‘which dog’. However, in (38a), there is
an additional referential noun ‘Hanako’ placed between the foci of the
question, which would otherwise be contiguous. This shifts the focus away
from the foci of the question, and for this reason, this is analogous to the
English examples in (40), where the actual focus is shifted away from the
extraction site to a semantically highlighted element. This argument is
supported by the observation that the wh-question in (38a) becomes per-
fectly acceptable if the recipient of the giving is not overtly expressed, as
shown in (44).
(44) [dare-ga kureta] inu-ga shinde-shimatta?

who-nom give.pst dog-nom die-asp.pst
‘The dog that who gave (Hanako, you, etc.) died?’
The claim above is supported further by the following data. As shown in

(45), broad-reading wh-questions become increasingly awkward, if there is
gradient acceptability for a speaker, as more referential expressions inter-
vene between the foci of the questions. The degraded acceptability correlates
with the degree of lexical-semantic highlighting on the intervening elem-
ents, which shifts the focus away from the foci of the question [dare-ga
okutta] hon ‘the book that who sent’ (¼ ‘which book’).
(45) a. [dare-ga okutta] hon-ga nakunatta no?

who-nom send.pst book-nom become.lost.pst fp
‘The book that who sent had been lost?’
b. ?[dare-ga taroo-ni okutta] hon-ga nakunatta no?
who-nom Taro-dat send.pst book-nom become.lost.pst fp
‘The book that who sent to Taro had been lost?’
c. *[dare-ga yuubinkyoku-de taroo-ni okutta] hon-ga nakunatta no?
who-nom post.office-loc Taro-dat send.pst book-nom become.lost.pst fp
‘The book that who sent to Taro at a post office had been lost?’

On the other hand, if the wh-expression is kept adjacent to the head of

the relative clause, the other focus of the question, the restriction
is ameliorated even if there are other referential expressions in the
relative clause. In (46b, c), the foci of the question represent a contiguous
focus unit; therefore, it is functionally equivalent to a single focus of
a question.
(46) a. [dare-ga okutta] hon-ga nakunatta no?

who-nom send.pst book-nom become.lost.pst fp
‘The book that who sent had been lost?’
b. [jon-ga dare-ni okutta] hon-ga nakunatta no?
Jon-nom who-dat send.pst book-nom become.lost.pst fp
‘The book that John sent to whom had been lost?’
c. [jon-ga sono yuubinkyoku-de dare-ni okutta] hon-ga nakunatta no?
Jon-nom that post.office-loc who-dat send.pst book-nom become.lost.pst fp
‘The book that John sent to whom at the post office had been lost?’
In the case of narrow-reading wh-questions, on the other hand, acceptabil-

ity is not affected by additional referential phrases in the relative clause.
Examples are given in (47b, c). The wh-question is not blocked in these cases
because the relative head noun does not represent a focus of the question
(i.e. what is to be identified by the question) and does not form a focus unit
of the question with the wh-expression.
(47) a. [dare-ga dasu] hon-ga yoku ureru no?

who-nom publish book-nom well sell fp
‘Books that who publishes sell well?’
b. [dare-ga amerika-de dasu] hon-ga yoku ureru no?
who-nom America-loc publish book-nom well sell fp
‘Books that who publishes in America sell well?’
c. [dare-ga amerika-de jerii-paaneru-to dasu] hon-ga yoku ureru no?
who-nom America-loc Jerry.Pournelle-com publish book-nom well sell fp
‘Books that who publishes with Jerry Pournelle in America sell well?’
Also, the preceding discussion is relevant to multiple wh-questions, as

shown in (48). Multiple wh-questions are functionally similar to broad-
reading wh-questions because both types of question contain double foci.
In both cases, the question becomes constrained if there are intervening
referential phrases, as is the case in (48b, c).
(48) a. dare-ga nani-o okutta no?

who-nom what-acc send.pst fp
‘Who sent what?’
b. ?dare-ga taroo-ni nani-o okutta no?
who-nom Taro-dat what-acc send.pst fp
‘Who sent Taro what?’
c. *dare-ga yuubinkyoku-de taroo-ni nani-o okutta no?
who-nom post.office-loc Taro-dat what-acc send.pst fp
‘Who sent Taro what at the post office?’

Again, if the foci of a question are kept contiguous as shown in (49b, c),
these questions are considerably more acceptable, if not perfectly
acceptable.
(49) a. dare-ga nani-o okutta no?

who-nom what-acc send.pst fp
‘Who sent what?’
b. taroo-ni dare-ga nani-o okutta no?
Taro-dat who-nom what-acc send.pst fp
‘Who sent Taro what?’
c. yuubinkyoku-de taroo-ni dare-ga nani-o okutta no?
post.office-loc Taro-dat who-nom what-acc send.pst fp
‘Who sent Taro what at the post office?’
To summarize, whereas wh-questions in complex sentences in Japanese are

widely acceptable, restrictions are observed in wh-questions that are func-
tionally complex with double foci. The underlying principle is the required
association of the assertion part of an utterance and a displaced entity (i.e. a
wh-expression in situ in the case of Japanese wh-questions); this means that
the displaced entity (or a wh-expression in situ) must function in the focus of
the utterance.
Before concluding, we consider a constraint on ‘why’ questions. It has
been pointed out that the wh-adverbial why exhibits some peculiar charac-
teristics across languages (Jin 2019). Likewise, it has been observed that naze
in Japanese exhibits wh-island effects unlike the other wh-words (Lasnik and
Saito 1984; Richards 2000; Fujii and Takita 2007). For example, while naze in
a complement clause is acceptable, as shown in (50a), a complex NP and an
adverbial clause are unacceptable with naze in (50b, c).
(50) a. jon-wa [taroo-ga naze kinoo sono hon-o katta tte] itta no?
John-top Taro-nom why yesterday that book-acc buy.pst comp say.pst fp
‘Did John say that Taro bought the book yesterday why?’
b. *jon-wa [taroo-ga naze kinoo katta] hon-o karita no?
John-top Taro-nom why yesterday buy.pst book-acc borrow.pst fp
‘Did John borrow the book which Taro bought yesterday why?’
c. *[taroo-ga naze kinoo sono hon-o yonda atode] shiken-ga
Taro-nom why yesterday that book-acc read.pst after exam-nom
umaku itta no?
well go.pst fp
‘Did the exam go well after Taro read the book yesterday why?’
On the other hand, the other wh-adverbials such as dooyatte ‘how’ and
nominal wh-adverbials such as {donna/dooyuu} riyuu-de ‘for what reason’ are
not subject to the restriction, as shown in (51).13
(51) a. jon-wa [taroo-ga kinoo dooyatte tsukutta] ryoori-o tabeta no?

John-top Taro-nom yesterday how make.pst dish-acc eat.pst fp
‘Did John eat the dish which Taro made how yesterday?’

b. [taroo-ga kinoo donna riyuu-de kaetta kara]

Taro-nom yesterday what.kind.of reason-by go.home.pst because
jon-ga okotta no?
John-nom get.upset.pst fp
‘Did John get upset because Taro went home yesterday for what reason?’
The contrast in acceptability between (50) and (51) suggests that the
observed restriction is due to the property of the particular wh-word why.
Van Valin (2002: 169) points out the adsentential nature of why and states
that the answers to who, what and how involve some change in the original
sentence (which is also the case with the nominal wh-adverbial for what
reason in (51b)), but the answer to a why question is normally a because-clause,
which is adjoined to the unchanged original sentence. Why by itself can be a
felicitous question when it follows a preceding statement, and the answer to
the question does not change or elaborate the preceding statement, as
shown in (52).
(52) A1: John was upset.

B: Why?
A2: John was upset because Taro was late.
As Van Valin (2005: 284) explains, the interclausal ‘reason’ relation, [LS1]
BECAUSE′ [LS2], contains two independent logical structures, which, how-
ever, are linked by a subordinating conjunction. Therefore, there is only one
illocutionary operator for the two logical structures. For this reason, only
one of the clauses can be the potential focus domain at a time.14 This is
exemplified by (52A2), in which ‘Taro was late’ is in the potential focus
domain and ‘John was upset’ represents the presupposition. In a why ques-
tion, the wh-word, which corresponds with [LS2] of the interclausal semantic
relation above, must be in the potential focus domain since it is the focus of
the question. This means that the clause representing the other logical
structure must be outside the potential focus domain, necessarily represent-
ing presupposition.15
Then, why do we observe the extraction restriction on questions such as
those in (50b, c)? There is a simple Gricean explanation for this. Since the
non-wh-part of a why question must be presupposed, only the presuppos-
ition that is relevant to the point of the question must be expressed,
because of the maxim of relevance. In (50b), for example, the presuppos-
ition ‘Taro bought the book yesterday’ is relevant to the why question since
it represents one of the logical structures in the ‘reason’ relation, but ‘John
borrowed the book’ is not relevant because it is not part of the semantic
relation. Inclusion of the irrelevant proposition in the question must draw
the focus for the Gricean reason, which shifts the potential focus domain
away from the wh-expression. This is essentially the same Gricean explan-
ation which applies to the unacceptable questions in (40) discussed earlier.
With the account above, the acceptable why question formed in an object
complement such as (50a) is expected because the embedded clause is a

direct daughter of the matrix clause node (see the principle in (6)). In other
words, what is expressed by the embedded clause is required by and hence
directly relevant to the matrix clause. To summarize, the fundamental
functional requirement that the wh-expression must function in the part
of an utterance which represents the assertion is a valid principle for naze
‘why’ questions in Japanese as well.
16.4 Conclusion
This chapter has focused on extraction restrictions in complex sentences,

which are typically discussed under the assumption that such restrictions
are structurally definable. This is largely due to the tradition of subjacency-
based accounts, long associated with extraction restrictions, as well as
English-type languages being used as the cardinal ground for analysis. As
discussed throughout this chapter, RRG combines structural and func-
tional constraints to capture both cross-linguistic and language-internal
variation in extraction restrictions. The fundamental functional require-
ment is the association of the assertion part of an utterance and a dis-
placed entity (or a wh-expression in situ) such that the displaced entity (or a
wh-expression in situ) must function in the part of the utterance which
represents the assertion. Obviously, the three extraction constructions
discussed in this chapter, wh-question formation, topicalization, and rela-
tivization, represent different functions. As Van Valin and LaPolla (1997:
627) put it, the wh-element of a wh-question represents a focus, the topic in
a topicalized sentence represents either a topic or a focus, and the head
noun of a relative clause represents a topic, hence the construction-specific
variations. Topicalization in English such as ‘that book I wouldn’t buy’
contains the preposed element in the pre-core slot, hence, it is focal. In
Japanese, a topic in the detached position is topical, and a contrastive
topic, which is placed within the clause, whether core-internally or exter-
nally, is focal, as discussed in Section 16.3.2.
Despite these different functions, however, they share the underlying
property that they must function in the assertion part of an utterance (i.e.
within the PFD). While there is an obvious connection between a wh-element
(which is focal) and assertion, relativization and topicalization share the
property that ‘the clause in which the displaced NP functions is always about
the referent of the NP’ (Van Valin and LaPolla 1997: 627). In other words,
such a clause functions as predication about the displaced element, and
‘predication and “assertion about” are fundamentally related notions’
(p. 629).
The analysis of Japanese shows that the RRG principle which relates
extraction to assertion is valid for the language, despite its radical difference
from other languages with respect to the possible range of PFD. This offers
an important implication for the universality of the principle, particularly

because it is assumed (Van Valin 1996, 1998, 2005; Van Valin and LaPolla
1997) that the constraints are ultimately derivable from Grice’s (1975)
cooperative principle and the maxim of quantity, which are the general
principles of rational human behaviour (hence, a challenge to the subja-
cency principle as evidence of autonomous syntax, and ultimately to a
theory of universal grammar). Van Valin (1998) uses the following from
Kempson (1975: 190) to illustrate how the Gricean explanation is rooted in
the claim that the focus domain is the domain for extraction (i.e. question-
ing and predication).16
The speaker believes the hearer knows (and knows that the speaker
knows) a certain body of propositions (i.e. that there is a pragmatic
universe of discourse) and in making a certain utterance . . . he believes
that the hearer, knowing the conventions of the language and hence the
conditions for the truth of the proposition in question, will recognize a
subset of those conditions as being part of that pragmatic universe of
discourse and hence neither assertible, deniable or queriable (without
violating the quantity maxim), and a second mutually exclusive subset
of the conditions as being outside the pragmatic universe of discourse.
This latter set, he will interpret as being asserted, denied, commanded
or queried.
In RRG, the cross-linguistic variations in extraction are captured in terms

of how deep into the sentence structure assertion may be represented. By
default, a matrix clause is always within the PFD; thus, it is a possible
extraction site across languages. In some languages, the PFD may be
extended to include those subordinate clauses which are tightly connected
to the matrix clause, as a direct daughter of a clause. As in the case of
Japanese, the PFD may be extended further to include other subordinate
clauses; yet, even in Japanese, detached topic positions are excluded from
the PFD. This exclusion is important corroboration of the RRG principle
since a topic does not represent assertion. In Kempson’s (1975) terms, a topic
represents part of the pragmatic universe of discourse which is neither
assertible, deniable nor queriable without violating the quantity maxim.
Extraction is allowed in a topic only if the topic is focal, hence contrastive,
and this conforms to the Gricean explanation. Lastly, the RRG principle is
corroborated further by the observed lexical-semantic influence on wh-ques-
tion formation in Japanese, which again relates extractability to the
Gricean explanation.
References
Allwood, Jens S. 1982. The complex NP constraints in Swedish. In E. Engdahl

and E. Ejerhed (eds.), Readings on Unbounded Dependencies in Scandinavian
Languages, 15–32. Stockholm: Almqvist & Wiksell.

Bickel, Balthasar. 1993. Belhare subordination and the theory of topic. In

K. H. Ebert (ed.), Studies in Clause Linkage, 23–55. Papers from the First Köln-
Zürich Workshop. Zürich.
Chaves, Rui P. 2013. An expectation-based account of subject islands and
parasitism. Journal of Linguistics 49: 285–327.
Chomsky, Noam. 1973. Conditions on transformations. In S. R. Anderson and
P. Kiparski (eds.), A Festschrift for Morris Halle, 232–285. New York: Holt,
Rinehart, and Winston.
Chomsky, Noam. 1986. Knowledge of Language. New York: Praeger.
Deane, Paul. 1991. Limits to attention: A cognitive theory of island phenom-
ena. Cognitive Linguistics 2: 1–63.
Deane, Paul. 1992. Grammar in Mind and Brain: Explorations in Cognitive Syntax.
Erteschik-Shir, Nomi. 1973. On the Nature of Island Constraints. PhD disserta-
tion, Massachusetts Institute of Technology.
Erteschik-Shir, Nomi. 2007. Information Structure: The Syntax–Discourse Interface.
Erteschik-Shir, Nomi and Shalom Lappin. 1979. Dominance and the func-
tional explanation of island phenomena. Theoretical Linguistics 6: 41–86.
Fujii, Tomohiro and Kensuke Takita. 2007. Wh-adverbials in-situ, their
island-(in)sensitivity and the role of demonstratives in wh-in-situ licensing.
Nanzan Linguistics: Special Issue 3(1): 107–126.
Grice, H. P. 1975. Logic and conversation. In P. Cole and J. L. Morgan (eds.),
Syntax and Semantics 3: Speech Acts, 41–58. New York: Academic Press.
Haig, John. 1979. What relative clauses are about. Papers in Linguistics 12:
57–109.
Haig, John. 1996. Subjacency and Japanese grammar: A functional account.
Hasegawa, Nobuko. 1981. A Lexical Interpretive Theory with Emphasis on the Role
of Subject. PhD dissertation, University of Washington.
Hasegawa, Nobuko. 1985. On the so-called ‘zero pronouns’ in Japanese. The
Linguistic Review 4: 289–341.
Hasegawa, Yoko. 1989. Questioning vs. identifying: A functionalist analysis
of the [a candidate that which professor recommended was hired?] con-
struction in Japanese. In K. Hall, M. Meacham and R. Shapiro (eds.),
Proceedings of the 15th Annual Meeting of the Berkeley Linguistics Society,
138–149. Berkeley, CA: Berkeley Linguistics Society.
Hofmeister, Philip. 2007. Retrievability and gradience in filler-gap depend-
encies. In M. Elliott, J. Kirby, O. Sawada, E. Staraki and S. Yoon (eds.),
Proceedings from the Main Session of the Forty-Third Annual Meeting of the Chicago
Linguistic Society: Vol. 43–1, 109–123. Chicago, IL: Chicago Linguistic
Society.
Hong, Sun-ho. 2003. On island constraints in Korean. In G. K. Iverson and
S.-C. Ahn (eds.), Explorations in Korean Language and Linguistics, 107–125.
Seoul: Hankook Publishing.

Inoue, Kazuko. 1976. Henkei bunpoo to nihongo (Transformational Grammar and

Japanese). Tokyo: Taishukan.
Jensen, Anne. 2001. Sentence intertwining in Danish. In E. Engberg-
Pedersen and P. Harder (eds.), Ikonicitet og struktur, 23–39. Preprint
from Netværk for Funktionel Lingvistik, Department of English,
University of Copenhagen.
Jin, Dawei. 2013. Information structure constraints and complex NP islands
in Chinese. In Stefan Muller (ed.), Proceedings of the 20th International Confer-
ence on Head-Driven Phrase Structure Grammar, 110–120. Stanford, CA: CSLI
Publications.
Jin, Dawei. 2019. A semantic account of quantifier-induced intervention
effects in Chinese why-questions. Linguistics and Philosophy 43: 345–387.
Kempson, Ruth M. 1975. Presupposition and the Delimitation of Semantics. Cam-
Kennedy, Becky. 1989. On Bridging. Unpublished manuscript, Harvard
University.
Kluender, Robert. 1990. A neurophysiological investigation of wh-islands. In
K. Hall, J.-P. Koenig, M. Meacham, S. Reinman and L. A. Sutton (eds.),
Proceedings of the 16th Annual Meeting of the Berkeley Linguistics Society,
187–204. Berkeley, CA: Berkeley Linguistics Society.
Kluender, Robert and Marta Kutas. 1993. Subjacency as a processing phe-
nomenon. Language and Cognitive Processes 8(4): 573–633.
Kuno, Susumu. 1973. The Structure of the Japanese Language. Cambridge, MA:
MIT Press.
Kuno, Susumu. 1987. Functional Syntax: Anaphora, Discourse and Empathy. Chi-
cago, IL: University of Chicago Press.
Kuno, Susumu and Ken-ichi Takami. 1993. Grammar and Discourse Prin-
ciples: Functional Syntax and GB Theory. Chicago, IL: University of Chicago
Press.
and the Mental Representations of Discourse Referents. Cambridge: Cambridge
University Press.
Lasnik, Howard and Mamoru Saito. 1984. On the nature of proper govern-
ment. Linguistic Inquiry 15: 235–289.
Matsumoto, Yoshiko. 1997. Noun-Modifying Constructions in Japanese: A Frame
Semantic Approach. Amsterdam: John Benjamins.
Miyagawa, Shigeru. 1987. Wa and the wh phrase. In J. Hinds, S. K. Maynard
and S. Iwasaki (eds.), Perspectives on Topicalization: The Case of Japanese wa,
Müller, Gereon. 1995. A-Bar Syntax: A Study in Movement Types. Berlin: De
Gruyter Mouton.
Nakagawa, Natsuko, Yoshihiko Asao and Naonori Nagaya. 2008. Information
structure and intonation of right-dislocation sentences in Japanese. Kyoto
University Linguistic Research 27: 1–22.

Nishigauchi, Taisuke. 1986. Quantification in Syntax. PhD dissertation, Univer-

sity of Massachusetts, Amherst.
Richards, Norvin. 2000. An island effect in Japanese. Journal of East Asian
Ross, John Robert. 1967. Constraints on Variables In Syntax. PhD dissertation,
Massachusetts Institute of Technology.
Shimojo, Mitsuaki. 1995. Focus Structure and Morphosyntax in Japanese: Wa and
ga, and Word Order Flexibility. PhD dissertation, University at Buffalo
(SUNY).
Shimojo, Mitsuaki. 2002. Functional theories of island phenomena: The case
of Japanese. Studies in Language 26: 67–123.
Shimojo, Mitsuaki. 2011. The left periphery and focus structure in
Japanese. In W. Nakamura (ed.), New Perspectives in Role and Reference
Grammar, 266–293. Newcastle upon Tyne: Cambridge Scholars
Publishing.
Simon, Mutsuko E. 1989. An Analysis of the Postposing Construction in Japanese.
PhD dissertation, University of Michigan.
Takami, Ken-ichi. 1992. Preposition Stranding: From Syntactic to Functional Ana-
lyses. Berlin: Mouton de Gruyter.
Takita, Kensuke. 2014. Pseudo-right dislocation, the bare-topic construction,
and hanging topic constructions. Lingua 140: 137–157.
Tomioka, Satoshi. 2009. Why questions, presuppositions, and intervention
effects. Journal of East Asian Linguistics 18: 253–271.
Van Valin, Robert D., Jr. 1996. Toward a functionalist account of so-called
extraction constraints. In B. Devriendt, L. Goossens and J. van der Auwera
(eds.), Complex Structures: A Functionalist Perspective, 29–60. Berlin: Mouton de
Gruyter.
mechanisms of language acquisition. In M. Tomasello (ed.), The New Psych-
ology of Language: Cognitive and Functional Approaches to Language Structure,
221–249. Mahwah, NJ: Lawrence Erlbaum.
and syntax. In E. Raxilina and Y. G. Testelec (eds.), Typology and Linguistics
Theory: From Description to Explanation, 511–524. Moscow: Languages of
Russian Culture.
guage 29: 161–175.
Yoshimura, Kyoko. 1992. LF subjacency condition in Japanese. Coyote Papers:
Working Papers in Linguistics 8: 143–162.

Notes
1 The term ‘extraction’ and related terms such as ‘wh-question formation’,

‘relativization’, ‘topicalization’ and ‘displaced element’ are used in the
present chapter only for expository purposes. Likewise, a gap in examples
is used only to show the place in which the displaced element functions.
RRG is monostratal and does not posit any movement operation to move
an element from one syntactic position to another.
2 Van Valin and LaPolla (1997: 487) note that the (c) response becomes
acceptable if the entire clause is repeated, as in ‘no, after she left’. This is
predicted by the RRG claim because the entire subordinate clause func-
tions as an information unit in this case.
3 However, ambiguity depends on the linear position of the wh-word. If it
intervenes with the subordinate elements, it is interpreted as modifying
the subordinate clause only (e.g. [kuroichekku-ga itsu kakunooko-ni iru-to]
sukinaa-ga itta?), and if it appears after the subordinate clause, it modifies
the matrix clause only (e.g. [kuroichekku-ga kakunooko-ni iru-to] itsu sukinaa-
ga itta?).
4 It has been claimed by Miyagawa (1987: 188) that the use of wa with a wh-
phrase is acceptable only under the following specific conditions which
represent explicit contrastiveness: an identifiable set of referents is pre-
supposed in the immediate conversational context and every member of
this set is exhaustively represented in the wh-wa question.
5 The negation given in (18B) is only mildly awkward, and this is predicted
by the contrastive property of the wa-marked topic, which contains a
focal element, as discussed later in this section.
6 See Shimojo (2011: 281–285) for discussion of marked use of the topic
marker, which involves a mismatch between the inherent focus structure
of the topic marker and the contextually determined focus structure of an
utterance. Such a mismatch yields a particular implicature due to focus
shifting caused by the inherent property of the topic marking.
7 The corresponding topicalization of (27b) is also perfectly acceptable,
which contrasts with the unacceptable topicalization counterpart of
(27a), which was given in (23).
8 The postposing considered here is the so-called non-pause-type postpos-
ing, in which the postverbal unit is prosodically part of the preceding
unit, produced with a single intonation contour and without an inter-
vening pause after the predicate. Prosodically separate postverbal units
are best considered an afterthought and are hence structurally separate
from the preceding unit (see Simon 1989 and Nakagawa et al. 2008 for
discussions of the two types of postposing).
9 For the PoCS-type postposing, the sentence may also be about the dis-
placed element, as in [ __ kiteiru] yoohuku-ga yogoreteiru yo ano shinshi-ga
‘That gentleman, the clothes that (he) is wearing are dirty’, in which the
nominative argument ‘that gentleman’ in the PoCS represents a focus.

10 The postposed argument ‘book’ in this example is outside the actual

focus domain but it is in the PoCS. The ‘book’ is not a topic of the
sentence since the matrix clause ‘X wants to meet the person’ is not
about the book, and for this reason, the ‘book’ is not topicalizable with
the topic marker wa in this sentence.
11 See Shimojo (2002: §4.2) for detailed discussion in this regard.
12 Takita (2014) calls the construction with a postverbal bare-topic pseudo
right dislocation and claims that bare topics are base-generated in this
construction, unlike standard right dislocation, which is generated by
‘repetition and deletion’.
13 Ambiguous wh-expressions such as nande ‘by/with what’ (nominal adver-
bial) and ‘why’ (adverbial) are island-sensitive only for the adverbial
interpretation (Fujii and Takita 2007).
14 See Bickel (1993: 33) for a related discussion about adsentential
subordination.
15 The presupposition requirement for a why question is also claimed by
Tomioka (2009: 265).
16 Van Valin (1998) shows that this analysis has important implications for
language acquisition. He shows that there is data relevant to the con-
straints in the input to children and therefore that they can be learned.
In essence, the critical construct of the RRG analysis is information
structure, which derives ultimately from the Gricean principle; there-
fore, an account of the acquisition of extraction constraints does not
have to appeal to a theory of universal grammar.

Part Four
Applications of RRG

17
Role and Reference
Grammar and
Diachronic Syntax
Ranko Matasović
17.1 Introduction
This chapter will briefly review the work on diachronic syntax and morpho-
syntax in Role and Reference Grammar (RRG) and show how the tools of
language description developed by this theory can be used to account for
several aspects of language change. Although RRG was designed primarily as
a synchronic syntactic theory, it has nevertheless been successfully applied
in description, and sometimes explanation, of a number of diachronic
syntactic processes (for a survey, see Wiemer 2008a). It is unsurprising that
most diachronic work in RRG was devoted to Indo-European languages, as
their history is generally better known than the histories of most non-Indo-
European languages, but there are exceptions to this claim. Among dia-
chronic studies dealing with non-Indo-European languages in the RRG
framework we can mention Ohori’s dissertation (1992) and article on the
development of Old Japanese clause linkage (1994) and Wilkins’ account of
the diachronic development of verbs expressing associated motion in
Mparntwe Arrernte, an Aboriginal language of Australia (1991). Ardis
Eschenberg’s dissertation on the article system of Omaha (2005) also con-
tains important insights into its diachronic development. Within the Indo-
European family, especially well represented are diachronic studies of con-
temporary languages such as German (Diedrichsen 2008), French (Kailuweit
2008), English (Nicolle 2008) and Croatian (Matasović 2012), but aspects of
the histories of dead languages such as Latin have also been analysed in the
RRG framework (Cennamo 2001).
Since RRG is not a theory that lays much emphasis on the intuition of the
ideal speaker/hearer as the source of evidence for the grammaticality of
sentences, and invented ‘laboratory’ examples do not play a great role in the
argumentation about the syntactic structures of a language, it is easy to
analyse dead languages with limited corpora within the RRG framework.
RRG was therefore successfully used in the analysis of some aspects of syntax

648 RANKO MATASOVIĆ
of such languages as Ancient Greek (Rupnik-Matasović 2008), Old English

(González Orta 2002) and Latin (Michaelis 1993). Although these studies are
chiefly synchronic and descriptive rather than diachronic, they are also
relevant for the development of historical syntax in the RRG framework. It
is clear that a reliable and careful description of the syntax of the early
stages of individual languages is a prerequisite for an account of the syntac-
tic developments that occurred between those early stages and their present-
day descendants.
The rest of this chapter is organized as follows: in Section 17.2 we
discuss a set of parameters by which different theories of diachronic
syntax can be compared and evaluated. We show how RRG answers the
fundamental questions about the nature and causes of syntactic change,
and argue that it is not a mere tool of linguistic description, but a theory
that makes falsifiable empirical predictions. In Section 7.3 we discuss
how syntactic change is represented in the RRG framework and how
different elements of syntactic structure are affected by diachronic devel-
opments. We also show how empirically attested types of syntactic
change can be analysed in RRG and argue that this theory allows for a
more fine-grained analysis of diachronic processes than theoretically
neutral accounts of diachronic syntax. Section 7.4 (Conclusion) summar-
izes the main points made in the chapter and proposes a few directions
for future research.
17.2 The Goals of Diachronic Syntax: RRG Compared

to Other Theories
A theory of diachronic syntax should be judged by the same scientific

standards as any other scientific theory, that is, it should conform to the
criteria of simplicity (or economy), empirical adequacy and predictive and
explanatory power. Different theories can thus be compared, assuming it is
agreed that, other things being equal, a theory is better than its rival if it is
as simple as possible and empirically adequate (not contradicted by evi-
dence) and if it also has more predictive and explanatory power than
its rival(s).
To be empirically adequate can mean two different things for a theory of
diachronic syntax, as it needs to be both typologically adequate (able to
analyse the structure of typologically different languages without imposing
any preconceived features on them) and diachronically adequate (able to
adequately represent which structures change and exactly how this happens
in attested or reconstructed diachronic processes). To describe a syntactic
change means to establish a set of rules leading from the syntactic repre-
sentation at stage A of a language L to another syntactic representation at a
subsequent stage B of the language L. It is clear that both syntactic repre-
sentations at stages A and B should be adequately described, so synchronic

Role and Reference Grammar and Diachronic Syntax 649
empirical adequacy is a prerequisite of diachronic empirical adequacy, and

RRG was specifically designed to be as typologically adequate as possible in
the analysis of languages of different types (Van Valin 2005).
The predictive and explanatory power of a theory reflects itself in its
ability to answer a number of empirically testable questions, such as the
following (Matasović 2002): (i) What exactly changes when syntactic struc-
tures change? (ii) Are there impossible syntactic changes, or are some
changes just more probable than others (and why)? (iii) What is the relation-
ship between syntactic, phonological, morphological, semantic and prag-
matic changes? (iv) What is the relationship between syntactic change and
language learning? Is change somehow tied to imperfect language learning,
or are there any other mechanisms of change? (v) What types of syntactic
change should be distinguished? Are there changes that are more ‘funda-
mental’ than others? Let us briefly sketch how RRG addresses these ques-
tions in turn.
i. The exact ‘locus’ of syntactic change directly depends on the architec-

ture and ontology of the syntactic theory. A change that involves a
simple rearrangement of elements in one theory can involve a radical
restructuring of different representations (or ‘projections’) in a different
theory. For example, in the history of French, the adjunct pas, which
originally meant ‘step’, was reanalysed as a negation operator. The
sentence Il ne voit pas la fille (he-neg.-sees-neg.-art.-girl) originally meant
‘He does not see the girl at all’, where pas was the equivalent of the
English adverbial expression at all. After the reanalysis (during the
Middle French period), the element pas lost its adverbial properties
and was reduced to the status of a purely grammatical morpheme,
essentially a negation always used in conjunction with the preverbal
negation marker ne. In the Chomskyan frameworks, we would represent
this change as a modification of labels on the constituent tree, but in
RRG a lexical element which was represented in the periphery of a
clause was reanalysed as an operator, which should be represented on
a different ‘projection’ (i.e. the operator projection rather than the
constituent projection).
ii. Chomskyan accounts of historical syntax (e.g. Lightfoot 1979, 2001;
Roberts 2007) explicitly deny the possibility of syntactic changes that
would affect the structure of Universal Grammar (UG), which is con-
ceived as being innate, syntax-driven and autonomous (independent of
other modules of grammar). RRG, as a theory that denies (or rather, does
not deem it necessary to posit) innate Universal Grammar makes no
such assumptions, which means that there are no a priori impossible
syntactic changes. However, some changes are rather unlikely if they
contradict certain functional principles that affect the structure of
grammars of human languages. This is another way in which RRG
(and other functional theories of syntax) differs from the generative

approaches, which often explicitly claim that every change not affecting
the UG is equally probable (Kroch 2001: 726). Thus, for example, in RRG
it would be natural to claim that demonstratives are more likely to turn
into definite articles than into aspect markers on verbs, both because of
their similar function (establishing reference and marking definite ref-
erence) and because they have similar positions in the layered structure
of the reference phrase (RP). Such a claim amounts to a statistically
verifiable prediction which can be checked by looking at the history of
various languages in which the diachronic path from demonstratives to
definite articles can be observed. For example, in the Romance lan-
guages the articles developed from Latin demonstratives ille and iste, in
Ancient Greek the Homeric demonstrative ho became an article by the
Classical period, and similar developments can be observed in the his-
tory of a number of Balkan languages (e.g. Bulgarian and Albanian) and
Armenian (for other examples, see Matasović 2002: 59). The development
from demonstratives to aspect markers on verbs is, to the best of our
knowledge, unattested.1 A theory that cannot or does not make such
verifiable predictions is clearly weaker with respect to its predictive
power than RRG.
iii. Syntactic theories that presuppose the autonomy of syntax would tend
to predict that syntactic change is generally independent of changes in
other domains of grammar. RRG, as a theory that explicitly denies that
syntax is autonomous (in the generativist sense), sees syntactic change
in close interdependence with changes in phonology, morphology,
semantics and pragmatics. For example, in a number of languages it
can be observed that the emphatic pronoun, used for emphasizing (or
contrasting) the reference of a personal pronoun, becomes the reflexive
marker, used to establish obligatory coreference in the clause or dis-
course. This change, which occurred in the history of the English
reflexive (him-self etc.) as well as in Irish (féin), Ancient Greek (he-autón)
and other languages, is motivated by the semantic and pragmatic prox-
imity between emphasizing that a referent controls the action (rather
than someone else) and asserting that a referent acts upon himself/
herself (rather than upon someone else). Since the emphasizing pro-
noun, by virtue of its meaning and pragmatic use, has to have an
antecedent, it can also be used to establish coreference between itself
and its antecedent. Thus, in many languages the emphasizing and
reflexive markers are homophonous, as in the English sentences John
saw himself in the mirror (reflexive himself) and John closed the mirror himself
(emphasizing). The reverse process (from emphasizing to reflexive) is
unattested, as far as we know, since the use of a reflexive marker to
emphasize that a referent controls the action would not be pragmatic-
ally or semantically motivated.
iv. Generativist theories of syntactic change tend to assume that change
essentially takes place during language learning and is a consequence of

imperfect acquisition of the mother tongue. On the other hand, func-

tionalist theories, including RRG, make no predictions as to the connec-
tion between language learning and syntactic change. In principle, it is
possible that syntactic change occurs during adulthood, and that it is
caused by several other factors (besides imperfect learning of the first
language), for example interference with other languages in bilingual
social groups, imitation of linguistically creative and socially prestigious
individuals, etc.
v. A lot of work in generativist diachronic syntax has been devoted to
showing that syntactic changes are S-curved by nature, proceeding grad-
ually until a setting of ‘parameters’ is affected, when they become
abrupt and far-reaching, basically changing the typological profile of a
language within a generation. Thus, while the parametric approaches to
diachronic syntax, coupled with the minimalist assumption of an
underlyingly universal head-first clause structure, have to predict that
the more common diachronic change will generally be from the OV
order to the ‘more economical’ VO order (Roberts 2007), RRG does not
make any such assumptions, and both of the attested changes – from OV
to VO and from VO to OV – are equally possible and ‘natural’ (see later in
this section). RRG does not posit any fundamental parameters from the
setting of which other features of a language’s syntax would follow, but
it accepts the insights of language typology that certain syntactic fea-
tures tend to cluster together in languages, at least statistically, and that
tendencies towards typological consistency can often be discerned in the
history of languages. There are usually functional explanations for
clustering of syntactic features in languages of a certain type, and
diachronic developments toward typological consistency can therefore
often be viewed as being motivated by a general principle of economy, or
ease of processing. Thus, it has been shown that Greenbergian word-
order universals can be explained by a general principle to the effect
that languages tend to be consistent in their branching direction, such
that non-lexical syntactic categories tend to consistently branch either
to the left or to the right – this is Dryer’s ‘branching direction theory’
(1992). In diachronic terms, this will mean that languages will tend to
change their syntactic structures in such a way as to increase consist-
ency in the direction of their branching categories. For example, the
order of the relative clause and the head noun it modifies will tend to be
harmonic with the order of the verb and its object, so that in languages
with OV order relative clauses will usually precede their head nouns,
while in VO languages they will follow them, since both relative clauses
and objects (as potentially complex RPs) are branching categories. It is
important to note that Dryer was able to show that not all of Green-
berg’s word-order universals really hold cross-linguistically, if a suffi-
ciently representative language database is constructed and areal and
genetic biases are excluded. For example, the position of adverbs with

respect to verbs is not correlated with other pairings of syntactic con-

stituents, and this is because both adverbs and verbs are non-branching
lexical categories. A theory that clearly distinguishes lexical and poten-
tially branching syntactic categories, such as RRG (Van Valin and LaPolla
1997: 70–71), is clearly easier to square with Dryer’s results concerning
word-order correlations, and it is also better at predicting those types of
syntactic changes that are leading towards a more consistent word
order, which is – according to the branching direction theory – easier
for processing.
There is no general consensus as to what constitutes explanation in

diachronic syntax, but in principle, every theory is expected to show how
the impossibility, or statistical rarity, of certain syntactic changes follows
from its general principles. A theory that assumes that VP is a universal type
of syntactic constituent, as most varieties of generativism do, will find it
difficult to explain why the VP does not behave as a syntactic constituent in
diachronic processes, that is, why we seldom find that a language of the SOV
type changed to OVS, or vice versa, while changes from VSO to SOV are
rather common (e.g. in Ethiopian Semitic languages), just like changes from
SOV to SVO (e.g. in Romance and Germanic languages). A theory like RRG
that explicitly denies the cross-linguistic universality of VP as a constituent
does not face this problem, since it does not predict that VP (or V and O as
single constituents) will behave as a unit in diachronic changes. This does
not mean that RRG would exclude the possibility of a change from SOV to
OVS, since there may exist diachronic paths of word-order change that
involve simultaneous movements of more than one constituent, but such
changes are likely to be much less common, cross-linguistically, than
straightforward changes involving just one of the non-controversial con-
stituents such as the verb or one of its arguments.
However, a theory that makes absolutely no predictions about possible and
probable diachronic change is less of a theory and more like a set of devices
for describing historical processes. It must be emphasized, therefore, that
RRG does make some predictions about the diachronic probability of syntac-
tic changes, and these predictions either follow from its architecture and
ontology or they can be precisely formulated using the conceptual tools
developed by the theory. Let us look at a few examples.
RRG claims that the order of operators in the operator projection is
universal, and predicted by the Natural Serialization Principle (NSP, cf. Foley
and Van Valin 1984: §5.3; Van Valin and LaPolla 1997: 51; Van Valin 2005: 12).
NSP predicts that the order of operators in languages is highly restricted, so
that operators of broader scope cannot be closer to the nucleus (the verbal
root) than operators with narrower scope. This means, for example, that
there is no language in which a sentence operator such as illocutionary
force is closer to the root than a nuclear operator, such as the morpheme

expressing the aspect of the verb. In a language that has only suffixes, this
will necessarily mean that the suffix for aspect is closer to the root than the
suffix expressing illocutionary force, for example the question marker. Now
a consequence of the universality of the NSP is that no syntactic change is
possible that would violate it, which means that, for example, if a language
already has suffixes expressing illocutionary force – or some other clause- or
sentence-level operators – no new suffixes expressing operators with a
narrower scope, such as aspect, can develop. Thus, in Indo-European lan-
guages that had developed tense markers expressed by suffixes, new aspect-
ual markers could not be grammaticalized as suffixes, but rather developed
as prefixes, for example in Old Irish, where the simple preterite car-ais (love-
3sg.pret.) ‘he loved’ is opposed to the new perfect ro-car-Ø (prf.-love-3sg.). In
other languages, the development was more complex, for example in Slavic,
where full aspectual oppositions developed only after the new prefixed
forms (e.g. Croatian po-gledati ‘take a look’) came into opposition with non-
prefixed (gledati ‘watch’) and both prefixed and suffixed forms (po-gleda-avati
‘watch repeatedly’), but in no Indo-European language can we observe a
development of new suffixes or postverbal particles expressing aspectual
meanings, as such morphemes would necessarily follow inflexional mor-
phemes expressing tense, thus violating the NSP (for details see Matasović
2002: 63–64 and Wiemer 2008a for some corrections).
The NSP plays a role in another generalization that has been suggested
concerning the frequency of grammaticalization patterns. It has been
argued that operators with narrower scope tend to grammaticalize as oper-
ators with broader scope, but not vice versa (Foley and Van Valin 1984:
216–217; Matasović 2008). This means that, for example, core operators such
as morphemes expressing deontic modality will develop into clausal oper-
ators such as those expressing status (epistemic modality), but that the
reverse pattern would be rare or unattested. An example of such a pattern
is the change of the English modal must from deontic (The letter must arrive
next week) to epistemic (The letter must be in the mail). Similarly, nuclear
operators such as those expressing aspect regularly grammaticalize as oper-
ators with scope over the clause or sentence, such as markers of evidenti-
ality. An example of this pattern is the development of the participle of the
auxiliary ima ‘have’ in Macedonian, which was originally just the marker of
perfective aspect, but it has developed into an evidentiality marker express-
ing the ‘hearsay’ meaning, for example in the sentence Ti si imal kupeno kniga
(you be.3sg.pret have.ptcp bought.ptcp book) ‘You bought a book (someone
told me)’. This empirical generalization, which certainly requires further
elaboration and empirical corroboration, is not directly implied by any
principles of RRG, but the NSP, which is one of the theory’s fundamental
principles, allowed its precise formulation. In theories which do not posit
the NSP or its equivalent, it would not be possible either to formulate this
generalization or to check it empirically in a systematic manner. However,

just as any other hypothesis formulated in the framework of a scientific

theory, it could eventually be refuted, if enough counterexamples
are discovered.
In a similar vein, it has been argued (Wiemer 2008b) that, at least in Indo-
European languages, the diachronic development seems to be towards a
tightening of the correlation between case assignment and privileged syn-
tactic argument (PSA) selection (see Chapter 7). In early Indo-European
languages, including Germanic, Greek and Balto-Slavic, we often find verbs
which do not take the accusative as the case of their lowest macrorole
argument, but in which that argument can still become PSA in some
constructions, for example the passive. In later stages of these languages,
the lowest macrorole arguments are always marked with the accusative.
Likewise, in early Indo-European languages we often find non-nominative
marked PSAs, for example in Old English, which allowed non-nominative
PSAs (Roberts 1995: 176f.).2
(1) ac gode ne licode na heora geleafleast ac asende

But God.dat.sg neg like.pret neg their faithlessness.nom but send.pret
Him to fyr of heafonum.
them.acc to fire of heaven.dat.pl
‘But Godi didn’t like their faithlessness, but Øi sent them fire from heaven.’
In (1), the dative-marked RP (gode) is the privileged syntactic argument of

the coordinative construction, as the coreferent RP was deleted in the
second coordinated clause. In Modern English, on the other hand, the
(pronominal) experiencer argument of the first verb would have to be in
the nominative case (he rather than *him). Actually, only Icelandic preserves
the pattern attested in Old English, which probably goes back to Proto-
Germanic. It remains to be seen whether such a trend towards tighter
correlation between case marking and PSA selection can be found in the
history of other language families for which we have sufficient evidence,
but RRG at least allows us to formulate such a hypothesis very precisely. On
a more general level, it has been claimed that PSAs (or pivots, as they used
to be called in earlier versions of RRG) generally tend to develop from two
sources: either from the semantic notion of macroroles (Actor and Under-
goer), or from the pragmatic notion of topic. The first scenario has been
assumed for languages such as English (and presumably many more Indo-
European languages), while the other development probably took place in
the Philippine languages such as Tagalog (Foley and Van Valin 1984:
134–148), in which PSAs still have some topic-like features (for example,
only PSAs can be relativized, and heads of the relative clauses may be seen
as their topics, i.e. as what the relative clauses are ‘about’). If this is correct,
in languages such as English, semantically motivated features such as case
marking, which had correlated closely with semantic roles, became grad-
ually de-semanticized and acquired purely syntactic functions. In lan-
guages of the Philippine type, on the other hand, pragmatically

motivated topic and/or focus markers gradually lost their pragmatic func-
tions and became markers of PSAs. It remains to be seen whether this
general diachronic typology of PSA development can be extended to other
languages and language families.
There may well be other directional and universal patterns of diachronic
development that are precisely formulable in the RRG framework, but the
current level of our knowledge is insufficient to make strong hypotheses. For
example, in a number of ancient or archaic Indo-European languages the
reflexive pronouns derived from the Proto-Indo-European root *s(w)e- can be
used in the embedded clause to refer back to its antecedent in the main
clause. This is the case with Latin se in (2).
(2) Pompeiusi a me petivit ut sei-cum et apud sei

Pompey from me ask.3sg.ind that with.himself and at himself
essem cottidie.
be.1sg.sbjv daily
‘Pompey asked me to be with him and at his house, daily.’
(Gildersleeve and Lodge 1992: 325)
In terms of RRG’s layered structure of the clause we can say that the
domain of reflexivization in such languages is the clause, since the reflexive
marker is within the same (superordinate) clause, although it is not the
argument of the same verb as its antecedent (i.e. it is not in the same core).
On the other hand, in most modern Indo-European languages (including the
Romance languages, which developed from Latin), the antecedent of reflex-
ivization and the reflexive marker (derived from *s(w)e-) must be in the same
core,3 which means that the domain of reflexivization is the core, not the
clause. For example, the reflex of Latin se in French, the clitic se, cannot be
used in the subordinate clause as coreferent with the subject of the main
clause.4 It appears, then, that the archaic use of the reflexive marker found
in Latin was changed in a way that led to the narrowing of the domain of
reflexivization in many modern Indo-European languages (with the excep-
tion of Icelandic, which allows for ‘long-distance reflexivization’ but seems
to be archaic in this regard, cf. Van Valin and LaPolla 1997: 613 and Wack-
ernagel 2009: 515 for ‘long-distance reflexivization’ in Old Icelandic). It
would be interesting to know if there are attested instances of diachronic
developments of reflexive markers by which the domain of reflexivization in
a language was broadened, rather than narrowed. This is clearly a worth-
while topic for future research.
17.3 Representing Syntactic Change in the RRG

Framework
Since any aspect of the syntax and the levels of representation that are
mapped to it in the linking are subject to change, representing diachronic

processes involves showing how they affect different elements of syntactic

representation and the linking algorithms. In RRG terms this means that
diachronic changes can affect:
• the inventory of syntactic templates

• the inventory and arrangement of operators
• the focus structure projection (the potential focus domain)
• the logical structure of verbs
• the linking algorithms (e.g. the case-assignment principles).
Representing changes in all the affected levels of representation gives us

a more fine-grained picture of historical processes than is usual in theory-
neutral accounts of diachronic syntax. To give an example, what would in
pre-theoretical terms be described as change in clause alignment from
ergative to nominative-accusative in northern dialects of Tabassaran
(Harris and Campbell 1995: 249–250), in RRG involves specifying how this
change affected all of the elements above. Tabassaran was originally an
ergative language, just like the other languages of its family (NE Cauca-
sian), with subjects of transitives in the ergative case and objects and
intransitive subjects in the absolutive case. However, at some stage, topics
were cliticized on verbs, and, as subjects were the most common topics,
this cliticization became obligatory for subjects. At that time, the clitics
still distinguished the ergative and the absolutive case of the subject.
Southern dialects of Tabassaran still preserve this stage (note that the
first-person singular pronoun has the same form in the absolutive and
the ergative).
(3) uzu gak’wler urgura-za

I.erg firewood.abs burn-I.erg
‘I burn firewood’
(4) uzu urgura-zu

I.abs burn-I.abs
‘I burn, am on fire’
In RRG terms, we would say that at this stage the following changes
occurred: (1) The inventory of syntactic templates was enriched with a
construction in which the topic was obligatorily copied in postverbal pos-
ition. The full pronouns were preserved in the pre-detached position in such
constructions (the default position for topics). (2) The focus structure was
affected, as the topical element now had to be postverbal. (3) By a subse-
quent change, the focus structure was affected again, as the clitics lost
topicality and became pure agreement markers expressing the lowest
macrorole argument of the verb (the absolutive argument) and the highest
argument of the verb (the ergative argument). This change affected the
syntactic templates of the language, as the clitics now became integrated
into the core.

Now, in the northern dialects, the verb-agreement system became entirely

nominative-accusative, as they lost the distinction between the two types of
verbal subject clitics. The form -za, which had originally been the first-person
singular ergative, was generalized (the case marking was also affected, but
this is a different issue).
(5) izu bisnu-za džaq’a

I.erg catch-I bird.abs
‘I caught a bird’
(6) izu t’irxnu-za

I.abs fly-I
‘I flew’
In RRG terms, this means that the linking algorithm (case assignment for
clitics) was affected at this stage, in that the rule that applies in them
states that the highest macrorole (clitic) argument received the
‘nominative’ case, just as in Latin or any other nominative-accusative
language, and unlike in Tabassaran’s closest relatives. This is only a step
from a further change in which the clitics became fully integrated into
verbal morphology as person endings agreeing with the ‘subject’ (the
highest-ranking macrorole argument) of the verb. In this system – which
is perhaps how contemporary Northern Tabassaran should be analysed –
the language presumably acquired the head-marking feature by which the
verbal suffixes, rather than free pronouns, should be interpreted as argu-
ments, and this again affected the inventory of syntactic templates in the
language. Similar developments can be observed in other languages, in
which pronominal clitics, originally used in pragmatically marked con-
structions to mark direct and indirect objects, became agreement markers
on verbs. This is what happened in several Balkan languages with ‘clitic
doubling’, including Macedonian and Bulgarian, and a similar process can
be observed in a number of Spanish dialects (especially in Argentinian
Spanish, see Belloro 2004). Although in those languages the change from
clitics in pragmatically marked constructions to agreement markers on
verbs did not lead to alignment change, the initial steps of the process
were very similar to what happened in Tabassaran, as the syntactic tem-
plates and focus structure were similarly affected. Thus, the RRG approach
allows us to clearly see the similarities and differences in historical pro-
cesses occurring in very different languages.
It is a reasonable prediction that complex changes affecting several
aspects of syntax will be cross-linguistically much rarer than simple changes
affecting only a single syntactic structure or representation. This would
mean that, for example, clause-alignment changes will be much rarer,
cross-linguistically, than simple word-order changes, which affect only the
syntactic templates of a language and not the other aspects of its syntax.
However, other factors having little to do with pure syntax have to be taken

into account, for example the discovery of areal linguistics that both word
order and clause alignment are areally stable features of language, which
are therefore unlikely to change except in some historically extraordinary
circumstances (for example mass migrations of speakers of a language to a
different language area). Thus, the Indian Subcontinent has apparently been
characterized as a language area where the SOV languages have predomin-
ated for at least two thousand years and possibly much longer (Vedic
Sanskrit, the ancestor of all Indo-Aryan languages, was SOV, and Proto-
Dravidian was also probably SOV). However, Romani, which is spoken by
people who emigrated from Northern India roughly a thousand years ago, is
SVO, like Armenian, Greek and most other languages it was affected by on
the path from India to Europe.
Synchronic generalizations about universal, or quasi-universal aspects of
syntactic structure always have their bearing upon hypotheses about pos-
sible and probable patterns of change. Thus, it has been noted that the
universal aspects of clause structure (the nucleus, core, periphery and
clause) are all semantically motivated, while the non-universal aspects (the
detached phrases, the extra-core slots) are pragmatically motivated (in the
sense that they are associated with constructions that have strong pragmatic
restrictions on their occurrence, Van Valin 2005: 8). It should be noted that
these are empirical findings, not something postulated by the theory, but
they have led to the hypothesis that semantically motivated features of
language are diachronically stable, while pragmatically motivated features
are rather unstable and more likely to change in the history of a language.
Although this hypothesis appears to be motivated by the theoretical archi-
tecture of RRG, it has not been proved or subjected to systematic research
beyond a rather impressionistic overview of the literature on grammaticali-
zation (Matasović 2002).
A theory of diachronic syntax must also be able to capture different types
of syntactic change. Instances of syntactic change can generally be classified
as one of three types: reanalysis, extension, and borrowing (Harris and
Campbell 1995). Reanalysis is a process that involves potentially ambiguous
syntactic structures which receive new structural representation and are
parsed differently by language speakers before and after the change in
question. Extension is the process of generalization of rules or constructions
from one domain to others, and borrowing involves the adoption of con-
structions from one language into another. All three of these types of change
can be represented in RRG, and their representation in the RRG framework
allows the researcher to make a more fine-grained analysis than is usual in
general accounts of historical syntax.
A case of syntactic reanalysis followed by extension is presented in the
RRG framework by Diedrichsen (2008). She shows how in colloquial German
the nuclear juncture involving the verb bekommen ‘get’ and a number of
ditransitive verbs of transfer was reanalysed as a passive complex verb form.
An example of such a passive construction is (7):

(7) Ich habe diesen Computer von meinen Eltern geschenkt

I have this.acc.sg Computer from my.dat.pl parent.dat.pl presented
bekommen.
got.ptcp
‘I was presented with this computer by my parents.’
Subsequently, this construction was extended to some other verbs with a

meaning similar to bekommen (e.g. kriegen and erhalten), so that in the con-
temporary colloquial language it is possible to form a passive with these
auxiliaries as well. Moreover, the construction was extended to a few mono-
transitive verbs (e.g. schneiden ‘cut’), so that a sentence such as (8) is now
acceptable.
(8) Er bekommt die Haare von ihr geschnitten.

he get.3sg.prs art.acc.pl hair.acc.pl from she.dat cut.ptcp
‘He gets his hair cut by her.’
Diedrichsen’s analysis allows us to see not only how reanalysis and exten-
sion work on the syntactic templates of the affected constructions, but also
how such syntactic changes slowly spread through the vocabulary of a
language, in a way that is affected by the lexical meanings of
individual items.
Borrowing of syntactic patterns can also be represented within the RRG
framework. Its existence and, indeed, widespread occurrence, do not pre-
sent a theoretical problem, since RRG does not make any theoretical
assumptions about when and how syntactic change takes place. As far as
the theory is concerned, syntactic change can occur both during language
acquisition, as a consequence of imperfect learning, and in adulthood,
either as a result of interference in the minds of bilingual speakers (syntactic
borrowing), or for some other reason. Here, again, the theory allows us to
represent different aspects of syntactic borrowing: firstly, borrowing can
involve the copying of specific syntactic templates as abstract patterns, for
example with respect to word order. This was the case when the SOV word
order was borrowed in Ethiopian Semitic languages such as Amharic from
neighbouring Cushitic languages such as Somali, or when the inversion rule
in the formation of yes/no questions was borrowed into French from Ger-
manic (Frankish), cf. French Vient-il? (comes-he) ‘Is he coming?’ with inver-
sion, which is also found in Dutch and German, but not in most
Romance languages.
Secondly, syntactic borrowing can involve the borrowing of individual
lexical items that has consequences for syntax, for example when a verb
such as German schmecken ‘to taste’ was borrowed into dialectal Croatian
(šmekati). Like German schmecken, which takes the dative experiencer and
the nominative source (e.g. EsNOM schmeckt mirDAT gut ‘It tastes good to me’),
Croatian dialectal šmekati has a quirky case frame (ToNOM miDAT dobro šmeka
‘It tastes good to me’). It remains to be seen whether such items can be
borrowed if the recipient language does not already have verbs with such

logical structures and case frames (Croatian has the inherited verb sviđati se
‘to like’, which has essentially the same syntax as the borrowed šmekati).
This also applies to instances where concrete forms are not borrowed but
the syntactic behaviour of individual lexical items is influenced by the
syntax of their translational equivalents in another language (syntactic
calques). For example, it is tempting to see the case frame of Croatian
darivati ‘present (imperfective)’ and darovati ‘present (perfective)’ as a syn-
tactic calque of Latin donare ‘present’, since both verbs show variable
undergoer selection, allowing both the theme argument and the recipient
to occur in the accusative (which is the default case of undergoers). This
means that two different thematic roles (with different positions on the
Actor-Undergoer Hierarchy) can be undergoers (Van Valin 2005: 111). The
Croatian verb darivati can take either accusative of the recipient and instru-
mental of the theme (darivati djecuACC kolačimaINS ‘to present children with
cakes’), or the dative of the recipient and the accusative of the theme
(darivati djeciDAT kolačeACC ‘to present cakes to children’), similarly to its Latin
equivalent which takes either dative and accusative (donare alicuiDAT
aliquidACC ‘to present something to someone’) or accusative and ablative
(donare aliquemACC aliquareABL ‘to present someone with something’; Latin
ablative is the usual equivalent of the Slavic instrumental case). However,
Croatian darivati has etymologically cognate verbs in other Slavic
languages, including Russian (darivat’), which also have two pairs of case
frames parallel to Croatian, which makes the hypothesis that those case
frames are inherited from Proto-Slavic more likely than the hypothesis that
they were independently borrowed from Latin. Inheritance seems even
more likely in light of the fact that verbs with such case frames are
otherwise almost unattested in Slavic: in Croatian, the only other example
is nuditi ‘offer’, which can also be construed with both the accusative and
the dative (nuditi djeciDAT kolačeACC ‘to offer the cakes to children’) and the
accusative and the instrumental (nuditi djecuACC kolačimaINS ‘to offer children
with cakes’). In the case of nuditi, a plausible explanation of its exceptional
case frame and variable undergoer selection properties would be that it is
the result of extension from darivati, which is inherited. The Latin verb
meaning ‘to offer’ (offerre) could not have been the model for nuditi, as it can
be construed only with the accusative and the dative (offerre aliquidACC
alicuiDAT ‘to offer something to someone’).
It seems fair to say that most lexical borrowing involves syntactic adapta-
tion to pre-existing patterns: for example, one does not expect to find cases
where verbs with ergative case frames are borrowed into a language with
nominative-accusative morphosyntax, or verbs showing person agreement
borrowed into a language with no person agreement. However, there are
documented cases of borrowing of lexical items that either introduce new
syntactic patterns into a language or contradict pre-existing ones. For
example, Tagalog, a genderless language, has borrowed adjectives from
Spanish, and those borrowed adjectives must agree in gender with their

head nouns in the RP (Schachter and Otanes 1972: 166–168; Matasović 2014);
the same rule holds for adjectives borrowed from Sanskrit into Marathi.
Conversely, Croatian, in which gender agreement in the RP is obligatory, has
borrowed a number of adjectives from English that show no gender agree-
ment (e.g. super ‘great, super’ and cool).
Although borrowing is easily accounted for and represented in RRG, it is
unclear whether there are constraints on grammatical borrowing in this
theory, and it remains to be seen if the theory offers any clues to such
potential constraints. For example, the borrowability of elements repre-
sented on the operator projection would need to be investigated. Research
into language contact has long shown that grammatical morphemes –
including inflexional elements – can be borrowed, though usually on a
smaller scale than lexical morphemes, but it is unclear whether there is a
difference of borrowability of different classes of grammatical morphemes.
Since RRG explicitly distinguishes two classes of grammatical elements with
respect to their syntactic representation, mapping some grammatical mor-
phemes in the constituent projection (e.g. causative markers, or person
markers of verbs), and others on the operator projection (e.g. aspect, tense
and illocutionary force markers), it would be interesting to investigate
whether these two classes differ with respect to borrowability. Again, it
seems that borrowing syntactic templates – which may be influenced by
pragmatic factors – is much more common than borrowing of elements
representable on the operator projection, which is strongly semantically
motivated. However, besides intuitions based on careful reading of litera-
ture on language contact, there is at present little systematic research to
corroborate such a claim.
Finally, in this chapter we have discussed the implications of RRG for a
theory of diachronic syntax, but RRG is actually more a general theory of
linguistic structure than a simple syntactic theory, since the interface of
syntax, semantics and pragmatics plays a central role in its architecture and
explanatory goals. Therefore, everything that has been claimed here about
historical syntax also has consequences for historical semantics and prag-
matics, and our discussion has shown a number of such consequences.
Moreover, the boundaries between morphology and syntax are set some-
what differently in RRG in comparison to other theories, so diachronic RRG
is also in part a theory of historical morphology. It has been established that
most bound grammatical morphemes were in origin independent words
with syntactic functions (‘Today’s morphology is yesterday’s syntax’, Givón
1971: 413), and this means that the change of their grammatical status will
often affect how they are represented in RRG. Thus, many Classical Greek
directional verbal prefixes (represented as nuclear operators in the operator
projection) were still independent adverbs in Homeric Greek, where they
would be represented as elements of the periphery in the constituent pro-
jection. For example, in Homeric Greek epı̀ ‘over, above’ is still separated
from the verb teínō ‘extend’, as in (9):

(9) Epı̀ nỳks tétatai deilo~ısi broto~ısin.

above night.nom.sg extend.3sg.prf.middle poor.dat.pl mortal.dat.pl
‘The night has extended itself over the weak mortals.’
However, by the Classical period (5th century BCE) we find it only as the
prefix in epiteínō ‘extend over, cover’ (Lindemann and Färber 2003: 47).
On the other hand, not all diachronic developments that involve a free
morpheme becoming bound need affect the way those morphemes are repre-
sented in RRG. For example, the clitic pronouns of Literary French, such as the
je ‘I’ (from Latin ego), have largely become bound person-marking prefixes in
colloquial French, but in both stages of its development je would probably be
represented as an argument in the constituent projection. Only if the gram-
maticalization process is taken a step further, and je becomes a pure person-
agreement marking on the verb, while the former topicalized stressed pro-
noun (moi) changes its status and becomes the pragmatically unmarked
subject pronoun (like English I or German ich), would the way the morpheme
je is represented change, as moi would be linked to the argument position in
the constituent projection (rather than being represented as the topicalized
element in the pre-detached position, as in (10)).
(10) Moi, je parle français.

I 1sg speak French
‘(As for me,) I speak French.’
17.4 Conclusion
It has been argued that Role and Reference Grammar provides useful tools
in the investigation of diachronic processes affecting the syntactic structure
of languages of various types. The fact that RRG is designed to be descrip-
tively adequate in representing typologically very diverse living languages,
and that its adequacy can be tested empirically on languages with unlimited
corpora and living native speakers should mean that this theory is also an
adequate tool for description of dead languages with limited corpora and no
native speakers. Describing two synchronic stages of a single language
within a uniform theoretical framework is a prerequisite for the objective
analysis of the changes that occurred from one stage to the other, and it has
been argued here that RRG is up to that task. We have also claimed that the
theory makes interesting predictions about the probability of certain types
of diachronic changes that can be tested empirically. Moreover, the archi-
tecture of the theory enables the investigator to formulate specific hypoth-
eses about the directionality of changes on particular levels of syntactic
structure, and we believe that fruitful applications of RRG lie in proposing
and testing such hypotheses, for example with respect to the diachronic
widening or narrowing of the scope of operators, the expanding or shrink-
ing domains of certain syntactic processes such as reflexive binding or

agreement, and the interaction of the focus structure domains with the
domains of purely syntactic processes. As the review of the published dia-
chronic applications of RRG testifies, this theory is also well designed to cope
with the description and explanation of cross-linguistically common
grammaticalization patterns.
Unfortunately, there has not been a lot of work in diachronic syntax
within the framework of RRG, but it is to be hoped that historical linguists
will realize the potential of this theory in the future.
References
Belloro, Valeria. 2004. A Role and Reference Grammar Account of Third-Person Clitic
Clusters in Spanish. Unpublished MA thesis, University at Buffalo (SUNY)
[available on the RRG website, https://rrg.caset.buffalo.edu/].
Cennamo, Michela. 2001. On the reorganization of voice distinctions and
grammatical relations in late Latin. In Claude Moussy (ed.), De Lingva
Latina Novae Questiones (Actes du Xème Colloque International de Linguistique
Latine, Paris-Sèvres, 19–23 avril 1999), 51–65. Louvain: Peeters.
Diedrichsen, Elke. 2008. The grammaticalization of the bekommen-passive in
a RRG-perspective. In Kailuweit et al. (eds.), 87–145.
Dryer, Matthew S. 1992. The Greenbergian word order correlations. Language
68: 81–138.
Eschenberg, Ardis. 2005. The Article System of Umonhon (Omaha). PhD disserta-
tion, University at Buffalo (SUNY) [available on the RRG website, https://
rrg.caset.buffalo.edu/].
Gildersleeve, Basil L. and Gonzalez Lodge. 1992. Latin Grammar. Edinburgh:
Nelson.
Givón, Talmy. 1971. Historical syntax and synchronic morphology: An
archaeologist’s field trip. Chicago Linguistic Society 7(1): 394–415.
Gonzalez Orta, Maria M. 2002. Lexical templates and syntactic variation: The
syntax–semantics interface of the Old English speech verb secgan. In Ricardo
Mairal Usón and Miguel J. Pérez Quintero (eds.), New Perspectives on Argument
Structure in Functional Grammar, 291–302. Amsterdam: John Benjamins.
Harris, Alice C. and Lyle Campbell. 1995. Historical Syntax in Cross-Linguistic
Perspective. Cambridge: Cambridge University Press.
Kailuweit, Rolf. 2008. Some remarks on RRG and grammaticalization:
French verbal periphrases. In Kailuweit et al. (eds.), 69–86.
Kailuweit, Rolf et al. (eds.). 2008. New Applications of Role and Reference Gram-
mar. Newcastle upon Tyne: Cambridge Scholars Publishing.
Kroch, Anthony S. 2001. Syntactic change. In Mark Baltin and Chris Collins
(eds.), Handbook of Contemporary Syntactic Theory, 699–729. Oxford: Oxford
University Press.

Lightfoot, David W. 1979. Principles of Diachronic Syntax. Cambridge: Cam-

Lightfoot, David W. 2001. How to Set Parameters. Arguments from Language
Change. Cambridge, MA: MIT Press.
Lindemann, Hans and Hans Färber. 2003. Griechische Grammatik. Teil II: Sat-
zlehre, Dialektgrammatik und Metrik. Heidelberg: Winter.
Matasović, Ranko. 2002. On representing syntactic change: Towards a
theory of diachronic syntax. Suvremena lingvistika 53–54: 57–72. http://
hrcak.srce.hr/16335.
Matasović, Ranko. 2008. Patterns of grammaticalization and the layered
structure of the clause. In Kailuweit et al. (eds.), 45–57.
Matasović, Ranko. 2012. Las construcciones reflexivas del croata desde una
perspectiva histórica. In Ricardo Mairal Usón, Lilián Guerrero and Carlos
González Vergara (eds.), El funcionalismo en la teoría lingüística: la Gramática
del Papel y la Referencia, 171–186. Madrid: Akal.
Matasović, Ranko. 2014. Adnominal and verbal agreement: Areal distribu-
tion and typological correlations. Linguistic Typology 18(2): 171–214.
Michaelis, Laura. 1993. On deviant case marking in Latin. In Robert D. Van
Valin (ed.), Advances in Role and Reference Grammar, 311–373. Amsterdam:
John Benjamins.
Nicolle, Steve. 2008. Scope and the functions of be going to. In Kailuweit et al.
(eds.), 58–68.
Ohori, Toshio. 1992. Diachrony in Clause Linkage and Related Issues. PhD
thesis, University at Buffalo (SUNY) [available on the RRG website,
https://rrg.caset.buffalo.edu/].
Ohori, Toshio. 1994. Diachrony of clause linkage. In William Pagliuca (ed.),
Perspectives on Grammaticalization, 135–149. Amsterdam: John Benjamins.
Roberts, Ian. 2007. Diachronic Syntax. Oxford: Oxford University Press.
Roberts, Linda. 1995. Pivots, voice and macroroles: From Germanic to uni-
versal grammar. Australian Journal of Linguistics 15(2): 157–214.
Rupnik-Matasović, Maja. 2008. Case assignment in Classical Greek. In Kailu-
weit et al. (eds.), 146–156.
Schachter, Paul and Fe Otanes. 1972. Tagalog Reference Grammar. Berkeley, CA:
Van Valin, Robert D., Jr. and LaPolla, Randy. 1997. Syntax: Structure, Meaning
Wackernagel, Jakob. 2009. Lectures on Syntax (edited and translated by David
Langslow). Oxford: Oxford University Press.
Wiemer, Björn. 2008a. Applications of RRG in diachronic syntax: Overview
and open questions. In Kailuweit et al. (eds.), 2–44.
Wiemer, Björn. 2008b. Changing relations between PSA-selection, macro-
roles and case assignment: Insights from the diachrony of Slavic, Baltic
and other Indo-European languages. In Kailuweit et al. (eds.), 157–203.

Wilkins, David. 1991. The semantics, pragmatics and diachronic develop-

ment of ‘associated motion’ in Mparntwe Arrernte. Buffalo Papers in Lin-
guistics 91(1): 207–257.
Notes
1 This does not mean that such a development is entirely impossible, only
that it is very improbable. Rare patterns of grammaticalization occasion-
ally lead to unexpected homonymy of grammatical morphemes. In
Omaha (a Siouan language) there is a set of articles which are also used
as auxiliary verbs and evidential markers, and Eschenberg (2005: 176–206)
convincingly shows how original copulas expressing orientation of their
subject were reanalysed as classificatory articles. For example, Webaxu khe.
Zi. (pencil be.horizontal yellow) ‘The pencil is positioned horizontally. It is
yellow’ was reanalysed as Webaxu-khe zi (pencil-definite yellow) ‘The pencil
is yellow’ (Eschenberg 2005: 183).
2 We use the Leipzig abbreviations (https://www.eva.mpg.de/lingua/resources/
glossing-rules.php) with the following addition: pret ¼ preterite.
3 Or the simple clause including the core and the pre-core slot (Van Valin
2005: 167).
4 I.e. one cannot say something like *Jeani pense que Pierre sei frappe (lit. Jean
thinks that Pierre himself is.hitting) ‘Jeani thinks that Pierre is hitting
himselfi’. French works exactly like English in this regard, but
unlike Latin.

18
Functional Acquisition
Processes in Child Language
Links to Role and Reference Grammar
Richard M. Weist
18.1 Introduction
This chapter reviews a functional theory of language acquisition based on

the proposal that children utilize their understanding of cognitive and
communicative principles to construct a grammar that integrates seman-
tic and pragmatic notions. The chapter explores child language data that
are relevant to such issues as how layered clause structure, operator
projection, predicate structure and grammatical relations are acquired
within this communication-and-cognition framework. Specifically, the
chapter explains how layered clause structure is founded on the cognitive
notions of predicate and argument. The cross-linguistic research on the
child’s emerging tense-aspect-modality system is related to issues of predi-
cate structure and operator scope. An evaluation of the semantic and
syntactic properties of grammatical relations is presented based on the
acquisition data. The chapter presents a brief description of some of the
basic assumptions of the functional theory of Role and Reference Grammar
(RRG, Van Valin 2005). In the process of showing how the language acquisi-
tion data map to the RRG framework, the chapter includes contrasts with
alternative theories, such as autonomous syntax theory. From the perspec-
tive of conceptual development, the infant-toddler is viewed as a relatively
proficient information processor with the capacity to discover fundamen-
tal linguistic relationships in the spirit of the theory of Operating
Principles (Slobin 1985).
The chapter begins with a presentation of related views of the semantic
structure of child language during the early multiple word phases of
acquisition (Section 18.2). This is followed by a minimal presentation of
the clause structure and operator scope proposed within the RRG frame-
work as well as the basic properties of predicate structure. With a focus on
Turkish, Japanese and Polish, the next section (i.e. 18.4) examines the role
of predicate structure in the acquisition of tense and aspect. The case for

Functional Acquisition Processes in Child Language 667
wide-scope processing of temporal, aspectual and modal information is

presented in Section 18.5, then Section 18.6 shows the importance of
restricted neutralization in the acquisition of grammatical relations, and
the concept of linking rules is related to child language data. Finally,
the acquisition of wh-questions is considered from different theoretical
perspectives in Section 18.7.
18.2 Semantic Relations in the Early Multiple Word Stage
The three research projects reviewed in this section share the following
properties: (1) they base their arguments on child language data, (2) they
propose that children understand predicate-argument structure that incorp-
orates semantic relations, (3) they do not find evidence to support a
Chomsky-type theory of the acquisition process (see, for example, Chomsky
1965 or Pinker 1984), and (4) they share with Van Valin (1993: 2), ‘the
conviction that grammatical structure can only be understood with refer-
ence to its semantic and communicative function’. However, none of the
investigators credit the toddler with the capacity to process inflectional
morphology (cf. the research on tense, aspect, and modality described in
Section 18.4).
Limited-Scope Formulae. In his seminal study of eleven children (20 to 26
months old) encompassing five different languages, Braine (1976: 4) pre-
sented the argument ‘that the first productive structures are formulae of
limited scope for realizing specific kinds of meanings’. According to Braine’s
observations, a limited-scope formula was ‘a rule that maps elements of a
semantic representation into positions in the surface structure’ (p. 69). Five
of the ten patterns summarized by Braine (pp. 56 and 57) are as follows, with
a semantic value linked to the position pattern: (1) identification, ‘it/that’ þ X;
(2) disappearance, ‘allgone’ þ X; (3) negation, ‘no’ þ X; (4) actor–action rela-
tions, Kendall swim, and (5) requests, ‘want/have-it’ þ X. Braine argued that
‘ALL early learning can be interpreted as a learning of position patterns’
(p. 66) and that the child language data do not support theories that
attribute grammatical structure such as S ! NP þ VP and VP ! V þ NP to
the language learner.1 In Section 18.6, I will return to an attempt to explain
the emergence of wh-question formation in English with a Braine-type
concept of ‘lexically-specific formulae’ (Rowland and Pine 2000: 157).
‘Case’ (or Semantic Function) Grammar. Bowerman (1973) investigated
the early stages of word combinations in two children learning Finnish,
Seppo and Rina, and one child learning English, Kendall. Citing Chomsky
(1965), Bowerman summarized that ‘the constituent structure assigned to
children’s utterances by transformational generative grammar is largely
gratuitous’ (p. 222) and specifically, ‘sentence-subjects tend initially to be
restricted largely to the semantic function ‘agent’ with a hand full of
exceptions for some children’ (p. 189). As an alternative, Bowerman

668 RICHARD M. WEIST
proposed a ‘case’ grammar based on the semantic functions agent (i.e. the
instigator of action) and object (i.e. the entity affected by action or state
identified) (see Fillmore 1968). In fact, Kendall and Seppo produced excep-
tions to the ‘agent-as-subject’ regularity, for example, Kendall’s pillow fell and
thread break and Seppo’s kissa putto ‘cat falls’, hirri pelä(sty) ‘mouse is-afraid’,
and torni kaatuu ‘tower falls-down’. Furthermore, the agent/subject is not
always animate, for example, helikopteri lentää ‘helicopter flies’. These ‘excep-
tions’ may preview the children’s understanding of more abstract grammat-
ical functions. The data indicate that children know more than ‘limited-
scope’ mapping rules (as Braine claimed), but the children were not yet
credited with the capacity to process morphosyntactic information (cf.
Toivainen 1980 for Finnish).
Usage-Based Grammar and Verb-Island Constructions. Tomasello (1992)
focused on the manner in which verbs emerged in the corpus of his daugh-
ter Travis from a cognitive-cultural perspective. According to Tomasello
(1992: 23), ‘young children’s early verbs and relational terms are individual
islands of organization in an otherwise unorganized grammatical system’,
and further, ‘In the early stages, the child learns about arguments and
syntactic marking on a verb-by-verb basis, and ordering patterns and mor-
phological markers learned for one verb do not immediately generalize to
other verbs.’ In contrast to the supposition that language emerges in the
child guided by innate universal grammar, Tomasello (2003: 21–31) pro-
posed that language is acquired by utilizing the principles of ‘intention-
reading’ and ‘pattern-finding’. According to the theory, children learn a set
of verb–argument constructions having a set of ‘slots’ to be filled with
specific semantic relations, for example, the schema ‘X made (this) Y’ was
realized as, Maria made this duck and Linda made ice. Before the age of 3,
Tomasello (2000: 71) found no motivation for productive verb categories or
inflectional morphemes. Hence, the verb in the construction [X singing]
would not be classified as an ‘activity’ verb, and the verb in [X fall-down]
would not be an ‘achievement’ verb. Further, neither the -ed in spilled Weezer
milk (Travis, 1;9) nor the -ing in Weezer drinking the eggs (Travis 1;7) were
judged to be productive morphemes coding tense and aspect because there
was very little evidence for contrast at this phase of development (cf. a rare
contrast: came off Grover (1;9) versus smoke coming out the chimney (1;11) (see
Tomasello 1992, Appendix). Contrary to Tomasello’s predictions, cross-
linguistic findings reveal the early influence of semantic relations on verb
categories and the rapid emergence of contrasts in the finite morphology
(see Tables 18.4 and 18.11).
Theories of Early Combinations and RRG. In his paper relating functional
linguistic theory to language acquisition, Van Valin (1991) presented an
argument for the relevance of RRG to child language research. Van Valin
began with the distinction between the ‘adaptationist’ and the ‘construc-
tionist’ views of language acquisition. Given the adaptationist view, the
child is innately programmed with a set of universal principles, and the

child sets parameters to conform to the target language (see, for example,
Pinker 1984, Radford 1990, or Sano and Hyams 1994). Alternatively,
according to the constructionist view, children learn language by utilizing
cognitive structures such as linguistic information processing principles
(see, for example, Slobin’s (1985) theory of ‘Operating Principles’). In their
analysis of single and early multiple word utterances, child language
researchers have observed the critical role that semantic and pragmatic
principles play in their data. The most fundamental principles of acquisition
that are derived from the child language data are also theoretical primitives
in the RRG theory. Thus, the RRG theoretical framework has the potential to
provide important insights into the question of how language is acquired.
This chapter will explore RRG’s potential in this regard.
18.3 Clause Structure, Operators and Predicate Structure in RRG
Clause Structure and Operators. The components of clause structure are

related to the scope of operators in Table 18.1 (cf. Van Valin, 2005: 12,
Fig. 1.4). Within the layered structure of the clause (LSC), the semantic units
motivate syntactic units at each layer of the structure, and grammatical
categories called ‘operators’ modify the clause structure at different layers.
In other words, the syntactic unit at a specified layer can be said to be within
the scope of a set of operators. Operators that modify the nucleus have a
narrow scope and those modifying the clause a wide scope. When I turn to
the child language data, I will consider the hypothesis that there is a
correlation between operator scope and the order of acquisition.
Semantic Structure of Predicates. Initially motivated by the work of
Vendler (1967) and Dowty (1979), Van Valin’s thinking about predicate
structure has been refined from the four basic categories found in Foley
and Van Valin (1984: 39, Table 2) through Van Valin and LaPolla (1997: 109,
Table 3.4 ) to Van Valin (2005: 45, Table 2.3). Table 18.2 contains the lexical
representations for the Aktionsart classes that are basic to the system and
are most frequently cited in language acquisition research. Predicates in the
four basic categories have the following properties: (1) States; static, (2) Activ-
ities; dynamic and atelic, (3) Accomplishments; dynamic and telic, and
(4) Achievements; dynamic, telic, and punctual. Predicates that contain
Table 18.1 Layered structure of the clause (LSC) and operator scope in RRG
Semantic units Syntactic units RRG operators in layers with scope
Predicate Nucleus Directional, Negation (Neg.), Aspect

Predicate þ Argument(s) (Arg.) Core Deontic modality, Directional, Neg.
Predicate þ Arg. þ Non-Arg. ¼ Clause ¼ Core þ Illocutionary Force (IF), Evidentials,
Proposition Periphery Tense, Status, Neg.
Proposition þ DP elements Sentence None

Table 18.2 Lexical representations for four basic Aktionsart classes
Aktionsart classes Examples Logical structure
State to be, know, want predicate′ (x) or (x, y)

Activity cry, sing, run around do′ (x, [predicate′ (x) or (x, y)])
Achievement break, fall, find INGR predicate′ (x) or (x, y)
Accomplishment learn Polish, melt, make toys BECOME predicate′ (x) or (x, y)
INGR (or instantaneous change) or BECOME (or change over time) in their
logical structure have the telic property. The course of language acquisition
is shaped by the interaction of the semantic structure of predicates with the
layered clause structure and its operators. Cross-linguistic research has
revealed a pervasive interaction of lexical and grammatical aspects. The
next few sections of this chapter will probe the child language data that
are relevant to this interaction revealing links to the RRG theoretical
framework.
Syntactic Realization of Logical Structure. Thematic relations are the
semantic roles/functions of the arguments in logical structure (see Van Valin
2005: 58, Fig. 2.3). The linking system to syntax involves two semantic
macroroles, Actor and Undergoer. These macroroles are generalizations
across argument types: Actor across agent-type and Undergoer across
patient-type thematic relations. The agent argument of DO is the least
marked choice for Actor, followed by the first argument of do′ (x), while it
is the most marked choice for Undergoer, while the patient/entity argument
of state pred′ (x) is the least marked choice for Undergoer, followed by the
second argument of pred′ (x, y), and the most marked choice for Actor. The
Actor-Undergoer Hierarchy specifies the mapping of thematically linked
arguments to macroroles, and, depending on language typology, mapping
continues to the privileged syntactic argument (PSA), which is the RRG
equivalent to syntactic subject (Van Valin, 2005: 127, Fig. 5.1). In a
nominative-accusative language like English, Polish or Japanese, the PSA
will be in the nominative case.
18.4 Predicate Structure and the Acquisition of Operators
Cross-linguistic research focusing on the acquisition of the finite morph-

ology of the tense-aspect-modality (TAM) system of child language has con-
sistently discovered a link between TAM morphology and categories of
Aktionsart as follows: (1) past tense – perfective aspect – telic Aktionsart
(e.g. achievements) and (2) non-past – imperfective aspect – atelic Aktionsart
(e.g. activities) (see, for example, Weist 1986, 2014a, b). The relationship
between the TAM domain of finite morphology and elements of predicate
structure is particularly theoretically revealing, as the evidence has been
found at a very early phase of language acquisition, and it extends across a

large variety of languages. From the RRG perspective, the acquisition data
provide insights into the relationship between the predicate that shapes
clause structure and the operators that modify levels of clause structure.
Predicate Structure and Operators in Turkish. I begin with a study by
Aksu-Koç (1998) on child Turkish because the investigation involved the
following four qualities: (1) it was sensitive to the emergence of TAM morph-
ology in a very young child (Deniz 1;3 to 1;11), (2) it included child-directed-
speech, (3) the classification of Aktionsart was empirically motivated, and
(4) the results revealed an unambiguous link to Aktionsart. Four multifunc-
tional affixes of Turkish were analysed, each having a value of tense, aspect
and modality (i.e. specifically evidential). The affixes with their meanings
and examples are as follows: (1) -dI past, perfective, and direct experience,
gel-dI ‘he came’; (2) -Iyor present and progressive, gel-Iyor ‘he is coming’; (3) -Ir
habitual-generic and possibility, gel-Ir ‘he comes’; and (4) -mIş past/present,
resultative, and indirect experience, gel-mIş ‘he has come evidently’. A set of
tests motivated by Dowty (1979) and organized for child language research
by Shirai and Andersen (1995) were used to categorize Deniz’s verbs into
categories of Aktionsart (i.e. state, activity, achievement and accomplish-
ment). Aksu-Koç (1998: 263 and 264) identified four stages of development as
follows:
1. 1;5.9 to 1;9.9: emergence and restricted use of -dI with achievement verbs
(e.g. bul- ‘find’)
2. 1;7.3 to 1;7.8: flexible use of -dI and emergence and restricted use of -Iyor
with activity (e.g. ye- ‘eat’) and state verbs (e.g. yat- ‘lie:down’)
3. 1;7.23 to 1;8.14: flexible use of -dI and -Iyor and emergence and restricted
use of -Ir and -mIş
4. 1;8.27 to 1;10.19: flexible use of all four forms.
Throughout these stages, the distribution of the target morphemes in

Deniz’s corpus closely resembled the distribution in her mother’s child-
directed speech. In her earlier research, Aksu-Koç (1988) studied the acquisi-
tion of the inflectional morphology of Turkish in three children: ES
(1;9–1;11), YK (1;11–2;1) and SÖ (2;2–2;5). A ‘restrictive stage’ was not found,
and the children were in the ‘flexible use’ phase during the earliest sample.
For example, in early samples, the three children inflected activity verbs
with -dI (i.e. past, perfective, direct) one-third less frequently than verbs
involving a change-of-state.
Predicate Structure and Operators in Japanese. Shirai (1993) investigated
the relationship between tense-aspect morphology and Aktionsart categories
in the corpus of Sumihare (Noji 1976, 1977, cited by Shirai). The research
focused on three tense-aspect morphemes: -ta (past), -te i- (durative), and -ru
(non-past). Regarding the interaction of tense-aspect morphology and Aktion-
sart in Japanese, the durative/continuative marker -te i- has an action-in-
progress meaning with activity and accomplishment verbs and a resultative
state meaning with achievement verbs. Within the Sumihare corpus, Shirai

Table 18.3 The percentage of verb tokens produced with -ta and with -te i- by
Sumihare for two predicate categories and three age periods
-ta -te i- (-teiru, -teru, -toru)
Aktionsart categories 1;6 1;7 1;8 2;1 2;2-1 2;2-2
Activity 11.1 7.9 28.3 81.8 52.6 47.9

Achievement 73.3 65.8 58.7 18.2 47.4 49.3
(1993: 200 and 201) analysed the emergence of -ta from 0;11 to 1;8 and -te i-
from 1;11 to 2;2. Table 18.3 contains the percentage of verb tokens produced
with -ta and with -te i- by the child Sumihare during three particularly salient
acquisition periods. For -te i-, the interval of 2;2 was split in half due to the
rapid development at age 2;2. The results show the relationship between
elements of predicate structure and tense-aspect morphology. At an early
age, the relatively high percentage of verb tokens were found in achieve-
ments with -ta (i.e. ‘past/perfective’) and activities with -te iþru (i.e. the action-
in-progress meaning). As the developmental process rapidly unfolded, the
likelihood of activity verbs with -ta increased, and the likelihood of achieve-
ment verbs with -te i- (i.e. the resultative state meaning) increased. As was
found with Deniz learning Turkish, the link between the do′ (x) predicate
structure and present-progressive meaning and between INGR pred′ (x) struc-
ture and past-perfective meaning was found in the Japanese child Sumihare.
However, like Aksu-Koç’s (1988) Turkish findings, there was no ‘restrictive’
stage since Sumihare inflected activity verbs as well as achievement verbs
with -ta.
Shirai (1998) continued his investigation of the relationship between
tense-aspect morphology (i.e. -ta, -te i-, and -ru) and categories of Aktionsart
with three more Japanese children: Aki (1;5 to 3;0) (Miyata 1995, cited by
Shirai), Yocchan (1;11 to 2;2) (Clancy 1985), and Taachan (1;10 to 2;2) (Kokur-
itsu, Kokugo and Kenkyujo 1982, cited by Shirai). When the tense-aspect
morphemes became productive, each child had a relatively high frequency
of verb types in one of the Aktionsart categories. For Aki, achievement -ta
verb types were the most frequent at 2;1 and remained highly frequent to
2;7, and activity-te i- verb types emerged as the most frequent at 2;4 and
remained so to 2;7, supporting Shirai’s (1993) investigation of Sumihare.
Aki’s mother’s child-directed speech was also analysed, and in general, the
relationship between tense-aspect morphemes and Aktionsart categories
supported Aki’s acquisition pattern. However, the ‘expected’ pattern was
not complete for Yocchan and Taachan. Following expectations, at product-
ive usage, the achievement category was closely linked to past -ta, but this
was not the case for durative -te i-. Hence, the expectations of Aksu-Koç’s
‘restricted stage’ are not always realized.
Predicate Structure, Operators and Finite Morphology in Polish. Weist
et al. (2004) utilized the corpora of six children learning Polish (1;3– to

4;11) and six children learning English (1;2 to 4;11) from the CHILDES
archives (MacWhinney and Snow 1985) to investigate the emergence of
tense-aspect morphology within the RRG theoretical framework. The
method of investigation was ‘predicate tracking’. Employing Dowty-type
tests, the process began by assigning verbs to four basic Aktionsart categor-
ies (i.e. state, activity, accomplishment and achievement). Following
Aktionsart classification, dynamic predicates were partitioned into telic
(i.e. accomplishment and achievement) and atelic (i.e. activity) sets (see
Table 18.2). Polish has perfective and imperfective aspect where imper-
fective is the least marked form, and English has progressive and non-
progressive aspect where the non-progressive (or ‘simple’) form is the least
marked. In Polish, there is an interaction of tense and aspect such that
non-past perfective verbs have future-tense meaning, while non-past
imperfective verbs have present-tense meaning. Imperfective future has a
periphrastic form. English has quasi-modal (i.e. going-to/will) future forms.
With two values of aspect and three values of tense, there might be six
inflectional forms to evaluate. However, present tense in perfective aspect
does not exist in Polish, and the children learning English in this study did
not use the future tense progressive aspect form. In order to balance the
cross-linguistic comparison, we reduced the statistical analyses to four
tense-aspect forms in each language:
1. Polish:
(a) perfective past and non-past (future meaning)
(b) imperfective past and non-past (present meaning)
2. English:
(a) non-progressive past and (going-to / will) future
(b) progressive past and present.
The average age of the acquisition of the tense-aspect morphology for

atelic and telic predicates in Polish and English is presented in Table 18.4.
The values in bold print refer to the telic data. Each value is calculated by
summing over the set of atelic and telic predicates for each child, and then
summing over the six children in the language group. Thus, Table 18.4
represents a relatively highly productive phase in the acquisition process.
The Polish data reveal an exceptionally clear interaction between the prop-
erty of telicity within predicate structure and grammatical aspect. If the
predicate is telic, perfective forms are acquired first, and in contrast, if the
predicate in atelic, the imperfective forms are acquired first. The underlying
form of predicate structure shapes the acquisition pattern. This lexical–
grammatical aspect interaction is also found in the English corpora, but
not as strongly. From the perspective of the RRG theoretical framework, it is
obvious that an adequate theory of language acquisition must include an
understanding of predicate structure (see Table 18.2), and predicate struc-
ture will have to have a primitive component within clause structure (i.e.
the nucleus for RRG; see Table 18.1).

Table 18.4 The average age of the emergence of tense-aspect forms for
(atelic/telic) predicates
(a)
Polish Aspect
Tense Perfective Imperfective
Past 3;4 / 2;2 2;7 / 3;1

Non-past (Present) Non-existent 2;1 / 2;8
*
Non-past (Future) 2;9 / 2;3 {2;5 / 3;0}
* {} The future imperfective data were scored but could not be compared to English.
(b)
English Aspect
Tense Non-progressive Progressive
Past 2;11 / 2;7 3;2 / 3;5

Non-past (Present) Excluded 2;10 / 2;10 *{2;4 / 2;7}
Non-past (Future) 2;11 / 2;8 Not observed
* {} ‘Present’ progressive forms were scored with and {without} the auxiliary.
The Weist et al. (2004) predicate tracking study evaluated three depend-
ent variables. In addition to the age of emergence (see Table 18.4), the
likelihood that a tense-aspect form would occur and contrasts in tense-
aspect forms were also investigated. The likelihood data complement the
emergence data; for example, a form that emerged early was highly likely.
Contrast is an important indicator of acquisition. Children acquiring a
relatively inflected language like Polish do not hear bare stems and neither
do they produce them. A tense and/or aspect morpheme may emerge as a
‘frozen’ form, and contrast provides evidence for functional status. Three
types of contrast were evaluated for all predicates (i.e. telic and atelic
combined) (1) tense only (e.g. past perfective / future perfective), (2) aspect
alone (e.g. past perfective / past imperfective), and (3) tense-aspect (e.g. past
perfective / present imperfective). The contrast findings are presented in
Table 18.5. Contrary to conventional wisdom, there was no evidence to
support the claim that aspect is acquired before tense. Furthermore, there
was considerable evidence that tense morphology codes deictic relations.
Weist et al. (2009) conducted a second predicate tracking study in order to
investigate the impact of predicate structure on the acquisition of
agreement and tense morphology. The data for this study were found in
the corpora of six children learning Polish (1;0 to 4;11) and six children
learning English (1;2 to 4;11) taken from a variety of sources (see, for
example, the Kraków project, Smoczyńska 1985). The set of predicates being
tracked were either stative or dynamic, with the dynamic set further
classified as atelic versus telic. The telic subset included predicates with

Table 18.5 The average age of the emergence of tense-aspect, tense, and aspect
contrasts summing over values of lexical aspect
Language Tense-Aspect Tense Aspect
Polish 2;10 2;4 2;11

English 3;0 3;0 3;3
Table 18.6 The percentage of agreement and tense contrasts for the three predicate
types in Polish and English
Language Polish English
Aspect PFV IPFV NPROG PROG
Type Agreement
Stative Impossible 100 100 Non-standard

Atelic 18 100 58 80
Telic 96 46 81 75
Type Tense
Stative Impossible 100 100 Non-standard

Atelic 6 100 95 65
Telic 97 34 98 35
BECOME/INGR in their logical structure. The single dependent measure was

contrast, and in one analysis, the contrast findings were evaluated with
grammatical aspect included, as shown in Table 18.6. Once again, the
impact of predicate structure as well as language typology is evident. In
both languages, contrasts in finite morphology are likely to be found in
dynamic predicates when bounded grammatical aspect is combined with
the semantic property of telicity. Regarding typology, the difference in the
behaviour of stative predicates is the most obvious. In Polish, the perfective
form of a stative predicate has inchoative meaning, making it telic (e.g.
lubić/ polubić ‘to like / to get to like’). In English, progressive aspect codes the
property ongoing, which is absent in the logical structure of a stative
predicate. The influence of language typology is also revealed in the acqui-
sition of dynamic predicates due to the fact that the marked aspectual form
in Polish is linked to external perspective (i.e. perfective (PFV)), and in
English the marked form is linked to internal perspective (i.e. progressive
(PROG)). In Polish, for example, the perfective member of an atelic aspectual
pair has delimited meaning, for instance, for pobawić/bawić ‘to play’ the
perfective pobawić means ‘to play for a while’ (see Weist et al. 2004: 50 for
examples). The delimited semantics may require additional learning (see
also Stoll 1998 for Russian).
Weist et al. (2009: 1335, Table 7) also found the average age and range of
the initial agreement and tense contrasts overall, that is, without

distinguishing values of grammatical aspect. In general, there was consider-

able overlap in the range of the acquisition of agreement and tense morph-
ology, that is, there was no evidence for a sequence (cf. Meisel 1994: 90,
where UG derives an agreement then tense acquisition sequence). Regarding
typology, agreement and tense morphology were acquired more rapidly in
the Polish children. Weist et al. (2009: 1329–1331) outlined the typological
differences between the English and Polish finite morphological systems,
and they show how information processing is facilitated within the Polish
system. Within the scope of Operating Principles theory, Slobin (2001: 441)
referred to this process as ‘typological bootstrapping’.
In summary, language acquisition theories based on limited-scope for-
mula, or slot-and-fill lexical concepts, or variations on these themes, have
no prospects (i.e. theoretical mechanisms) to explain the emergence of
contrasts in finite morphology which are paired with communicative com-
petence. Utilizing their capacity to process linguistic information, children
gain access to agreement and tense-aspect distinctions at a relatively early
phase in development. The evidence within discourse contexts demonstrates
the children’s understanding of the relevant contrast (e.g. 2nd versus 1st
person and past versus future tense).
18.5 Clause Structure and Operator Scope: Two Hypotheses
‘Layer-emergence’ Hypothesis. The relationship between the layers of clause

structure and the scope of operators within the RRG framework is outlined
in Table 18.1. Applying this structural relationship to the language acquisi-
tion process, it might be argued that operators modifying ‘simple’ layers of
clause structure should be acquired before operators modifying ‘complex’
layers (see Van Valin 1991). The argument went as follows: ‘Hence the status
of an operator as nucleus, core, or causal, which is a function of its inherent
semantic complexity, is the source of the increasing “cognitive complexity”
that Weist and Stephany claim to be the explanation of the order of acquisi-
tion’ (Van Valin 1991: 18).2 Since aspect modifies the nucleus and tense
modifies the clause, aspect morphology should emerge in child language
before tense morphology: the child should code temporal contour before
temporal relations between speech time and event time. I will refer to this
expectation as the ‘layer-emergence’ hypothesis. The early research on
aspect and tense in child language appeared to support this prediction;
see, for example, Aksu-Koç (1988), Antinucci and Miller (1976) and Bronckart
and Sinclair (1973). These investigators observed the emergence of finite
morphology with the potential to code both temporal contour (i.e. com-
pleted) and temporal relations (i.e. event time prior to speech time) (e.g. -dI
(past-perfective-direct) for Turkish). Regarding the child language data, it
was claimed that past-tense morphemes were never found with the
following properties: (1) inflecting activity verbs, (2) combined with

imperfective aspect, and (3) making reference to remote prior events. With
deference to Piaget’s theory of conceptual development (Piaget 1954, 1971),
the child language researchers ruled out the deictic function of past-tense
morphology, for example, ‘the child lacks an abstract conception of time
that would allow him to construct, between any two events X and Y, the
relation Event Y precedes Event X, even when no other type of relation holds
between event X and Y’ (Antinucci and Miller 1976: 184). Combining longi-
tudinal observations with experimental findings, Weist et al. (1984) dis-
covered that none of the three claims about the emergence of past-tense
morphemes were correct for child Polish. During spontaneous caregiver–
child interactions and within controlled elicitation conditions, 2-year-old
Polish children produced activity verbs inflected for imperfective aspect in
past tense referring to moderately remote prior events. In fact, the Polish
findings are supported by the Turkish (Aksu-Koç 1988) and Japanese (Shirai
1998) data, revealing a low frequency of activity predicates inflected for past
tense (see Table 18.3 for Japanese). Regarding the Piagetian conceptual
limitation argument, decades of research with infants and toddlers has
demonstrated that young children have the capacity to remember past
experiences (see, for example, Bauer 1996, 2007 and Rovee-Collier 1997).
The idea that the concept of aspect is functional in child language prior to
the concept of tense may also be influenced by a methodological artefact.
Child language data have often been collected in a ‘here-and-now’ restricted
environment; see, for example, Aksu-Koç (1998: 261) ‘During the sessions
Deniz and her mother played with toys, read, looked at pictures in books,
and talked.’ In contrast, Bowerman (1981: 1 and 2) took ‘copious notes’ on
her daughter’s utterances throughout the day, and she found that, ‘Spon-
taneous reference to both past and future (and nonactual) events was well
established from the one-word period.’ Some of Bowerman’s observations of
Christy’s temporal utterances are found here in Table 18.7. Christy’s capacity
to talk about the ‘not-here-and-now’ is quite consistent with the develop-
mental research on memory processes. In other words, children who can
think about prior and subsequent events can also talk about these events if
given the opportunity. Table 18.7 shows that when Christy was producing
uninflected two- and three-word combinations, she was locating event time
prior-to, at, and subsequent-to speech time. As the inflectional morphology
emerged, her use of tense matched the utterance context. Regarding past
time utterances, the Aktionsart of her verbs was atelic as well as telic.
Relating to the ‘layer-emergence’ hypothesis, while wider-scope operators,
such as tense, incur greater linguistic complexity, the more recent develop-
mental findings indicate that young children have the capacity to process
this information (see also Table 18.5).
‘Operator-Scope’ Hypothesis: Temporal Domain. Within the RRG frame-
work, the concept of scope defines the relationship between operators and
layers of clause structure. The ‘operator-scope’ hypothesis predicts that
children understand scope relations as the TAM morphology emerges.

Table 18.7 Melissa Bowerman’s observations of her daughter Christy’s

temporal utterances
Tense-aspect morphology absent
Time Utterance Context
Past (1;5). Write Sissy. She notices where her sitter had written on her hand.
Present (1;6) Kiddicar ride. Said while playing with the kiddicar.
Future (1;9) Pie open later. She anticipates taking an unopened pie home.
Tense-aspect morphology present
Past Activity (2;0) I cried. Remembering a walk earlier in the day.

Past Achievement (2;0) I didn’t fall Remembering an earlier incident in a wading pool.
down water. Emily did.
Past (2;0) Balloon was outside. Yesterday, her balloon was outside.
Present (1;10) Mommy Ø coming now. Said as her mother arrives at the door.
Future (2;1) She’s won’t get wet. Said holding a doll while sitting on the potty.
Regarding the temporal domain, as children acquire the concept of gram-

matical aspect, we expect to find evidence for a link to the concept of
nucleus, and as the concept of tense emerges, the relationship between
the concept of tense and the clause layer of structure will be found.
Regarding the link between the aspect operator and the nucleus layer, the
evidence for the link is decisive (see Table 18.4). The predicate tracking
research shows that predicates with the property of telicity in their semantic
structure, for example achievements, will emerge initially in perfective
aspect in Polish and non-progressive in English, and predicates lacking this
property (i.e. activities) will emerge first in the imperfective aspect in Polish
and progressive in English. This kind of operator to nuclear layer relation-
ship has been found cross-linguistically.
When tense morphology emerges, is there additional evidence that chil-
dren understand a link to clause structure? The argument can be made that
this link can be found in the relationship between tense and temporal
adverbs. According to Van Valin and LaPolla (1997: 169), ‘adverbs are like
operators, in that they modify different layers of the clause’. More specific-
ally, temporal adverbs are viewed as core modifiers. Thus, if temporal
adverbs emerge in child language during the same phase of acquisition as
tense, we will have evidence for a link to layers of clause structure. Weist
and Buczowska (1987) studied the emergence of time-related adverbs in
Polish. Considering the data from three children in the age range 2;4 to
3;2, Weist and Buczowska analysed utterances containing adverbs within
their discourse context. Table 18.8 contains the age of the initial occurrence
of each of the nine adverbs in at least one of the three children. The set of
adverbs was partitioned into three categories: immediate, cyclic, and
remote. Cross-linguistic research has shown that when deictic adverbs spe-
cifying a daily cycle emerge, they are likely to enter into mismatches with
tense. In Wawrzon’s sentence Jutro miałem ponton a już si˛e zepsuł ‘Tomorrow,

Table 18.8 The initial occurrence of the temporal adverbs in three sets of adverbs
Immediate Cyclic* Remote
Adverb Age Adverb Age Adverb Age
już 2;4 wczoraj 2;7 dawno 2;9

‘already’ ‘yesterday’ ‘long ago’
teraz 2;4 dzisiaj 2;5 kiedysˊ 2;11
‘now’ ‘today’ ‘in the past’
zaraz 2;5 jutro 2;7 później 2;9
‘soon’ ‘tomorrow’ ‘later’
* The initial occurrence of these adverbs always involved at least one tense-adverb
deictic contradiction.
(I) had (a) pontoon and already it broke’, the adverb jutro is pointing to the
future, and the tense (-ł-) is pointing to the past. From the perspective of
cognitive development, Friedman (1978: 281) argued that such incongruities
signalled a deficiency in the child’s early tense system. However, to the
contrary, the Polish data showed that children initially coordinate the
‘immediate’ temporal adverbs (już ‘already’, teraz ‘now’, and zaraz ‘soon’)
with tense. The tense–adverb incongruities emerge at a later phase of
acquisition when children are learning the meaning of the ‘cyclic’ adverbs.
The child’s earlier understanding of tense remains intact. While a direct
comparison of the emergence of temporal adverbs with tenses was beyond
the scope of the Weist and Buczowska study, the Polish data cited in
Table 18.4 indicate a close proximity. Relating to the ‘operator scope’
hypothesis, the Polish data support the claim that children demonstrate
an understanding of different layers of structure when tense-aspect morph-
ology is acquired.3 The aspectual adverb już ‘already’ modifies the nuclear
layer and the temporal adverbs teraz ‘now’ and zaraz ‘soon’ (not to mention
the cyclic and remote adverbs) modify the core layer.
‘Operator-Scope’ versus ‘Layer-Emergence’ Hypothesis: Modal Domain. In
the temporal domain, the ‘layer-emergence’ argument proposes that mor-
phemes with the potential to code deictic relations (i.e. tense) in the adult’s
language code temporal contour relations (i.e. aspect) in the child’s language.
Within the RRG framework, this argument is motivated by the fact that
aspect modifies a more primitive layer of clause structure than tense. There
is a companion argument in the modal domain (Van Valin 1991: 17, Fig. 3).
Since the operator ‘modality’ (covering deontic modality) is a core operator
and status (including epistemic modality) is a clause operator, along with
evidentials, deontic modality should be acquired before epistemic or eviden-
tial modality. Considering data from children learning Greek and English,
Stephany (1986) presented the argument that ‘Epistemic modal meanings
develop later than deontic ones in language acquisition’ (p. 393) and further-
more, the reason for this acquisition sequence can be found in the Piagetian
view of cognitive development. According to Stephany (1986: 393), ‘Studies of

Table 18.9 Sentence examples in the acquisition of existential meaning in Korean
Age Morpheme / knowledge Child’s initials and context Sentence
1;9 -TA new, direct TJ just finished a visual Chaca-ss-TA

search Find-PAST -TA
‘(I) found (it)’
1;11 -E assimilated PL overhears a conversation Polami meli ippukey hay-ss-E
about her recent haircut Hair prettily do-PAST -E
‘Had hair done, it was pretty’
2;2 -CI shared PL recalls having seen a Kweymwul Tibi-ey-na iss-CI?
monster on TV Monster TV-LOC -only exist-CI?
‘Monsters exist only on TV?’
2;4 -TAY indirect TJ talks to mother about a TV Sayka ppai ha-n-TAY
talking bird Bird bye do-PRESENT -TAY
‘Bird says “bye”’
cognitive development have shown that the notion of possibility as a

distinct form of reality develops in Piaget’s pre-operational stage (from
about 2 or 3 to about 7 or 8).’ Just as Piaget’s claims about memory
processes during the sensorimotor stage were deficient, so too were his
claims about the child’s conceptual capacity during the ‘pre-operational’
stage (see, for example, Mandler 1990; Newcombe and Huttenlocher 2003;
Spelke 1991).
Choi’s (1991, 1995) investigation of verb-final suffixes in Korean is relevant
to the operator-scope hypothesis. However, the child Korean findings do not
map directly to the Stephany (1986) and Van Valin (1991) proposed deontic
to epistemic acquisition sequence. Choi evaluated the acquisition of desire
and intention modal suffixes (cf. deontic, obligation and permission (Van
Valin, 2005: 9: Table 1.2)) and evidential modals. While the prototypical
deontic to epistemic sequence can’t be evaluated, the very early function
of the evidential modals (i.e. clausal operators like tense) can be evaluated.
Choi analysed the emergence of four evidential morphemes (-TA, -E, -CI and
-TAY) and two other morphemes (-LAY desire and -KKEY intention)4 in the
corpora of three children between the ages of 1;8 and 2;11. The evidential
morphemes have the following meanings: (1) -TA: new knowledge acquired
through direct experience and being assimilated, (2) -E: previously assimi-
lated knowledge, (3) -CI: previously established, certain, and shared know-
ledge, and (4) -TAY: knowledge learned from an indirect source. Table 18.9
contains sentence examples together with the utterance context revealing
the nature of the knowledge. The evidential morphemes became productive
prior to the desire and intention morphemes. The Korean data demonstrate
the robust nature of the infant-toddler’s information processing capacity
and their ability to understand the semantic complexity at different layers
of clause structure.

18.6 Grammatical Relations
Restricted Neutralization. Does child language contain grammatical rela-

tions, and if so, what kind? From the RRG perspective, it depends on the
language. While semantic roles are universal, syntactic functions are not,
making it very implausible that they are innate. According to Van Valin and
LaPolla (1997: 274), ‘grammatical relations exist only where there is a
restricted neutralization of semantic or pragmatic relations for syntactic pur-
poses’. Languages can be classified based on the nature of their restricted
neutralization possibilities, as summarized in Table 18.10. For typological
distinctions, S refers to the single argument of an intransitive verb, A refers
to the Actor of a transitive verb, U refers to the Undergoer of a transitive
verb, and d-S refers to, ‘the single NP of an intransitive verb derived via voice
from a transitive verb’ (Van Valin and LaPolla 1997: 268–269).
Returning to Section 18.2, in her investigation of the acquisition of
Finnish and English, Bowerman (1973) did not think that there was enough
evidence for restricted neutralization to justify or motivate the concept of
subject. While Kendall produced utterances like pillow fall, thread break, the
utterances of this form were infrequent and viewed as part of the ‘hand full
of exceptions’ to the ‘agent-as-subject’ rule. According to Tomasello (1992,
2003), item-based schemas included one or more ‘slots’ and a verb. The slots
were filled by semantic functions (see also Section 18.2). Once again, the only
‘grammatical’ relations were semantic in nature, supporting the assertion
that semantic roles are universal. Following Van Valin’s argument, we need
to find evidence for restricted neutralization in the child language data to
make the case for syntactic relations. Hence, Bowerman, Tomasello, and
Braine (as well) would compare child language to Acehnese and not to one of
the target languages, for example, English.
As discussed in Section 18.4, Weist et al. (2009) used the predicate tracking
methodology to discover how finite morphology emerges in children learn-
ing Polish and English in the age range from 1;0 to 4;11. By tracking the
emergence of this morphology within a set of predicates within each indi-
vidual child, it is possible to determine how the concept of restricted
neutralization develops. Within each language, agreement contrasts emerge
at a similar age across predicate types (see Table 18.11). In general, the
Table 18.10 Typology of restricted neutralization of semantic roles (Van Valin and
LaPolla 1997: 269)
Intransitive Transitive Grammatical ‘Subject’

Language verbs verbs relations classification
Acehnese no no No Semantic
Warlpiri yes no Yes [S,A]
English yes yes Yes [S,A,d-S]
Dyirbal yes yes Yes [S,U,d-S]

Table 18.11 Average age and range of initial person agreement contrasts in English
and Polish
English Polish
Predicate types Age Range Age Range
Stative 3;2 2;7–3;8 2;1 1;11–2;3

Atelic 3;1 2;5–3;9 2;4 2;0–2;7
Telic 3;1 2;4–3;4 2;4 2;1–2;6
acquisition process is more rapid in Polish, the more highly inflected language.
Limiting the analysis to Polish, we investigated the children’s capacity to
transfer their knowledge of agreement from one tense to a second tense. We
calculated the number of months it would take a child to make the transfer (i.e.
lag time), and we found that for some of their predicates, all of the children
transferred their capacity to form contrasts in agreement from one tense to
another within one month (i.e. zero lag). By approximately 2;4, restricted
neutralization is a property of child Polish, but exactly how do we classify
child Polish at this phase of development? At 2;4, it appears that child Polish
should be classified as an [S,A] type like Warlpiri at least. However, like English,
Polish has passive voice, and the Weist, et al. (2009) study did not evaluate this
possibility. This possibility was the focus of a paper by Weist (1990).
Weist (1990) investigated all of the facets of restricted neutralization in
child Polish, including the concept of voice. The research included the cor-
pora of four Polish children with ages ranging from 1;7 to 1;11 for two
children and 2;1 to 2;6 for the other two children. The children’s TAM system
was productive with contrasts in tense, aspect and mood. Transitive verbs
were sometimes observed with a direct object, and intransitive verbs never
were. The children contrasted nominative with accusative case, and their
verbs agreed with the privileged argument in person, number and gender.
During the period from 1;6 to 2;0, the children produced intransitive verbs
with a theme or patient as the core argument (i.e. with the macrorole Under-
goer) (e.g. spasˊ ć ‘to fall down’, przyjsˊ ć ‘to come/arrive’, and uderzyć si˛e ‘to get
hit’), and they produced intransitive verbs with an agent core argument (i.e.
with the Actor macrorole) (e.g. ˊspiewać ‘to sing’, płwać ‘to swim’, and bawić si˛e
‘to play’). Regarding the pseudo-reflexive particle si˛e (refl), the children
followed the rules for si˛e placement as follows: (1) immediately after the
verb: Marta (1;8) kr˛eci si˛e ‘(it) turns’, and (2) optionally, after the first stressed
element in the sentence: Marta (1;10) co si˛e kr˛eci? ‘what is turning?’ versus
Marta (1;9) czemu kr˛eca si˛e kóleczka? ‘why are the little wheels turning?’. In
Polish (child as well as adult), the transitive/intransitive contrast is distinct-
ive, as the following observation demonstrates: Kubusˊ (2;4) Otwiea-m. Tak si˛e
otwiera-Ø ‘(I) am opening (1sg) (the box). (It) opens (3sg) like this’. Here it is
clear that Kubusˊ has shifted from the transitive verb otwierać to the intransi-
tive otwierać si˛e. The transitive verb otwierać is inflected for first-person

singular, the pronoun is dropped, and the direct object is omitted. The
particle si˛e with the intransitive verb otwierać si˛e has been moved to the
position after the word tak, and the verb is inflected for third singular
agreeing with ‘it/box’. In summary, there is clear evidence children have
the capacity to neutralize the semantic roles Undergoer and Actor for
syntactic purposes. However, these observations demonstrate neutralization
that is restricted to the intransitive context. Polish children produce passive
participle verb forms in adjectival constructions (e.g. Bartosz (1;11) Zepsute, to
zepsute jest ‘Broken, (it) is broken’), but they do not produce passive voice
constructions (see Weist 1990: 1338–1341). Hence, child Polish has the prop-
erties of an [S,A] language (like Warlpiri) and not the fluent [S,A, d-S] lan-
guage (like adult Polish).
Linking Rules. Does child language contain grammatical relations, and if
so, what kind? While there are variations on these themes, there have been
two major approaches to answering this question. One approach begins with
the investigation of child language data and derives a hypothesis about
grammatical relations from what children produce and comprehend (see,
for example, Brown 1973; Bowerman 1973). The second approach begins
with a linguistic theory and makes predictions about what should be found
in the child language data that correspond to that theory (see, for example,
Pinker 1984; Radford 1990). Invariably, the investigators that begin with the
child language data arrive at the conclusion that grammatical relations are
learned, and those that begin with a linguistic theory (for example, some
version of Chomsky’s thinking) assume at the outset that grammatical
relations are innate. A revealing demonstration of these two alternatives
concerns Pinker’s concept of innate linking rules, and Bowerman’s observa-
tions of language acquisition in her daughters Christy and Eva. According to
Pinker (1984), children are innately equipped with linking rules that initiate
the process of mapping thematic roles onto syntactic functions. A hierarchy
of thematic roles and a hierarchy of syntactic functions were proposed to
establish the linking process. The hierarchy for thematic roles contains
agent, theme/patient, and location/goal/source, and the synchronized hier-
archy for syntactic functions contains subject, object and oblique object, in
that order. Direct links between thematic roles and syntactic functions (e.g.
agent to subject and patient to object) are ‘canonical’ (i.e. the innate
default), and the theory predicts that learning will be facilitated. The set
of thematic relations is defined by verb-argument structure. For an activity
predicate like to cry, the logical structure contains a single agent core
argument, and the agent is canonically linked to subject. An achievement
predicate like to fall has a single theme core argument, and since there is no
agent in the predicate structure, the theme is canonically linked to subject.
Non-canonical mapping occurs when the default links are ‘crossed’; for
example, the predicate to have contains the core arguments of location and
theme, where the location role becomes subject and not the higher-ranking
role of theme. For verbs taking three arguments such as to give, the canonical

Table 18.12 The acquisition sequence for canonical and non-canonical

constructions
Two-argument verbs
Child Age Canonical theme ¼ subject Non-canonical location/goal/source ¼ subject
Christy 1;9 Baby fall down (on) me Daddy have cake

Eva 1;8 Necklace stay (in) purse Ernie got spoon
Three-argument verbs
Child Age Canonical: Agent – verb – Non-canonical: Agent – verb – goal – theme
theme – goal
Christy 2;0 I gave a nipple Mommy I give Mommy a bottle

Eva 1;10 Christy bring that for (¼to) me Linda gave my (¼me) choochoo
mapping is as follows: agent to subject, theme to direct object, and benefi-

ciary to oblique object. The acquisition of sentences with the theme linked
to the oblique object should be delayed. Bowerman (1990) tested Pinker’s
hypothesis with observations of her daughters Christy and Eva, and some
critical comparisons are presented here in Table 18.12. When one observes
the child language data, evidence for innate linking rules is absent: Pinker’s
‘non-canonical’ constructions emerge at the same time as his ‘canonical’
constructions.5 In addition to anomalies found in the acquisition of English
(i.e. the child language data), there are other intractable predictions. Chil-
dren learning Acehnese will be at a loss to determine what to do with their
innate potential facing a target language void of syntactic relations, and
children learning ergative languages, such as Dyirbal, will experience gen-
etic interference as the target language links patient to subject and not
agent (see Van Valin 1992 on ergative languages and implications for lan-
guage acquisition).
18.7 Acquiring wh-questions
Rowland and Pine (2000) reviewed previous research on the acquisition

of the subject–auxiliary errors in the acquisition of wh-questions. They
pointed out that there is a phase when children produce un-inverted wh-
questions (e.g. What you did say?) and inverted yes/no questions (e.g. Does the
kitty stand up?) at the same time as they are beginning to produce correctly
inverted wh-questions (e.g. What did you doed?). The authors evaluate two
movement-rule accounts of the acquisition phenomena which they judged
to be lacking. Their alternative theory is based on the idea that children
have ‘a distributional learning mechanism that learns and reproduces
lexically-specific formulae’ (p. 157) and, more specifically, ‘lexically-specific
wh-word þ auxiliary combinations’ (p. 157). Rowland and Pine (2000: 163)

view their acquisition theory as a variation on Braine’s (1976) theory of

limited-scope formulae (see Section 18.2). According to the theory, children
learn ‘wh-word þ auxiliary markers’ which are highly frequent in child-
directed speech. In other words, there was no explanation for the underlying
dynamics that might foster an un-inverted wh-question. Rowland and Pine
analysed the corpus of Adam (see Brown (1973) with a focus on the ‘unin-
version period’, i.e. mean length of utterance (MLU) 3.24–4.10), and they
examined maternal utterances from a subset of transcripts prior to this
period. As auxiliaries entered Adam’s corpus, he began to produce both
inverted and un-inverted wh-questions, and during the ‘un-inverted period’,
the relative number of un-inverted versus inverted questions increased and
then decreased, terminating in a higher percentage of inverted wh-ques-
tions. The authors categorized every wh þ aux combination as inverted or
un-inverted, and they counted the frequency of occurrence in the maternal
speech sample, all well-formed. Some of Adam’s inverted combinations (e.g.
what do, where did) were frequent in the input, but others were not found in
the input (e.g. what was, which does). To make predictions worse, some of
Adam’s un-inverted combinations (e.g. why did/didn’t/don’t) were, in fact,
present in the input. No explanation was given for theoretical ‘misses’ (i.e.
zero frequency inverted, what was, and relatively high frequency un-
inverted, why don’t).
Van Valin (2002: 165) demonstrated that a general syntactic principle
combined with an understanding of information processing can predict
the acquisition of wh-question formation in Adam’s corpus. The principle
(the IF principle) is as follows: ‘Illocutionary force in English is signaled by
the position of the tense-bearing morpheme: a) In declarative utterances,
tense appears core-internally, and b) In interrogative utterances, tense
appears core-initially.’ In order to apply such a principle, the child needs to
be able to process tense (see Tables 18.4 and 18.5). The auxiliaries is, does and
has reveal their finite properties when they enter into the following con-
trasts: (1) am/are/is versus was/were, (2) do/does versus did, and (3) have/has versus
had. The tensed properties of the modals can, could, shall, will, may and might
are not as salient. This combination of the IF principle and morphological
processing brought Van Valin to the following prediction: ‘In wh-questions,
children will initially place only those auxiliaries which are explicitly tensed
in core-initial position.’ Hence, we can predict which wh-questions will be
inverted (i.e. is, are, was, do, does, did, have, has and had) and which ones will be
un-inverted (i.e. can, could, may, might, shall, should and will). Furthermore,
since the negative auxiliaries end in n’t, which does not signal tense, negative
auxiliaries will not be inverted in the initial phase of wh-question acquisition.
Van Valin’s predictions were based on the assumption that children process
wide-scope operators and integrate that information into the clause struc-
ture. The predictions based on this assumption were supported by the data
from Brown’s Adam provided by Rowland and Pine (2000). Van Valin (1998)
penetrated further into the acquisition of wh-questions showing how such

issues as the emergence of subject and object wh-question and constraints on

long-distance wh-questions can be understood within the RRG framework
without the need to propose an autonomous, parameterized Language Acqui-
sition Device or Universal Grammar.
18.8 Conclusion
In his paper linking RRG to language acquisition, Van Valin (1991: 9) pre-
sented two ‘perspectives’: (1) ‘Syntax is . . . relatively motivated by semantic,
pragmatic, and cognitive concerns’ and (2) ‘the child actually learns language
and CONSTRUCTS a grammar during the process of language acquisition’.
Regarding how the construction process functions, Van Valin cited Slobin’s
(1985) theory of Operating Principles. According to Slobin (1985: 1158), chil-
dren are innately equipped with a ‘language-making capacity [LMC]’ and
Slobin proposed that ‘LMC must begin life with some initial procedures for
perceiving, storing, and analyzing linguistic experience, and for making use
of capacities and accumulating knowledge for producing and interpreting
utterances’. The theory of Operating Principles has been criticized by child
language researchers, for example Bowerman (1985: 1281), and linguists, for
example Pinker (1989: 463), since Slobin’s information processing theory
lacks a model of grammar. This chapter was designed to investigate the
potential that RRG might be able to fill this void.
Semantic functions such as agent and theme/patient are operational in
the earliest phase of language acquisition, when single-word utterances
dominate the child’s corpus. The child’s capacity to process the finite
morphology of the target language becomes apparent as multiple-
morpheme utterances emerge, especially in highly inflected languages like
Turkish and Polish. As soon as there is evidence for the acquisition of the
elements of a tense-aspect-modality system, the semantic structure of the
predicate influences the pattern of acquisition. Hence, on the one hand,
children are utilizing their innate information processing capacity (which
might be referred to as their LMC), and on the other hand, this capacity to
process information is combined with at least some knowledge of predi-
cate structure (see Tables 18.3 and 18.4). While these two components of
linguistic knowledge are necessary to explain the acquisition data, they are
not sufficient. Children integrate this knowledge within a concept of
clause structure (see Table 18.1). Potential explanations of language acqui-
sition based exclusively on principles of information processing (e.g. Slobin
1985), or with categories of predicate structure (e.g. Shirai 1998) will fall
short if they fail to explain how integration into clause structure is accom-
plished. From the perspective of RRG, clause structure is created from the
semantic principle of predicate-argument structure, and it contains levels
of semantic complexity. The ‘operator-scope’ hypothesis presented here
predicts that children identify the scope of operators as the concepts of

tense, aspect, and modality are being acquired. The predicate tracking data
(e.g. Weist et al. 2004) showed that Polish children acquire tense (i.e. a
clause operator) and aspect (i.e. a nuclear operator) during the same phase
of acquisition. Similarly, Choi’s study of the emergence of evidential mor-
phemes (i.e. clause operators) found early productivity of clause operators
in the modal domain of the TAM system. These facts require that children
are constructing their understanding of clause structure simultaneously
with their decoding of the elements of the tense-aspect-modality system.
Van Valin’s (2002) study of the acquisition of subject–auxiliary inversion
revealed how the integration of the tense operator within clause structure
explains the emergence of wh-questions in child language data. From the
viewpoint of cognitive development, the child language data do not sup-
port the idea that the infant-toddler is somehow limited in their capacity
to process complex relations and construct meaningful representations in
the manner imagined by some early child language researchers such as
Braine (1976) or more recent researchers such as Rowland and Pine (2000,
2003). In contrast, the child language data reviewed here are consistent
with Spelke’s ‘core knowledge systems’ theory of cognitive development,
where the core knowledge systems are thought of as ‘mechanisms for
representing and reasoning about particular kinds of ecologically import-
ant entities and events’ (Spelke 2000: 1233). In retrospect, Slobin’s (1985)
set of Operating Principles might be thought of as the core knowledge
system for language.
References
Aksu-Koç, Ayhan. 1988. The Acquisition of Aspect and Modality: The Case of Past
Reference in Turkish. Cambridge: Cambridge University Press.
Aksu-Koç, Ayhan. 1998. The role of input vs. universal predispositions in the
emergence of the tense-aspect morphology: Evidence from Turkish. First
Language 18: 255–280.
Antinucci, Francesco and Ruth Miller. 1976. How children talk about what
happened. Journal of Child Language 3: 167–190.
Bauer, Patricia. J. 1996. What do infants recall of their lives? Memory for
specific events by one- to two-year-olds. American Psychologist 51: 29–41.
Bauer, Patricia. J. 2007. Remembering the Times of Our Lives: Memory in Infancy
and Beyond. Mahwah, NJ: Lawrence Erlbaum.
Bowerman, Melissa. 1973. Early Syntactic Development. Cambridge: Cambridge
University Press.
Bowerman, Melissa. 1981. Notes for the Nijmegen Workshop. MPI, Nijmegen,
Netherlands.
Bowerman, Melissa. 1985. What shapes children’s grammars? In Dan I.
Slobin (ed.), The Crosslinguistic Study of Language Acquisition, Vol. 2,
1257–1319. Hillsdale, NJ: Lawrence Erlbaum.

Bowerman, Melissa. 1990. Mapping semantic roles onto syntactic functions:

Are children helped by innate linking rules? Linguistics 28: 1253–1289.
Bronckart, Jean-Paul and Hermina Sinclair. 1973. Time, tense, and aspect.
Cognition 2: 107–130.
Braine, Martin D. S. 1976. Children’s first word combinations. Monographs of
the Society for Research in Child Development.
Braine, Martin D. S. 1992. What sort of innate structure is needed to ‘boot-
strap’ into syntax? Cognition 45: 77–100.
Brown, Roger. 1973. A First Language. Cambridge, MA: Harvard University Press.
Choi, Soonja. 1991. Early acquisition of epistemic meanings of Korean:
A study of sentence-ending suffixes in the spontaneous speech of three
children. First Language 11: 93–120.
Choi, Soonja. 1995. The development of epistemic sentence-ending modal
forms and functions in Korean children. In Joan Bybee and Suzanne
Fleichman (eds.), Modality in Grammar and Discourse, 165–205. Amsterdam:
John Benjamins.
Chomsky, Noam. 1965. Aspects of a Theory of Syntax. Cambridge, MA: MIT Press.
Clancy, Patricia. 1985. The acquisition of Japanese. In Dan I. Slobin (ed.), The
Crosslinguistic Study of Language Acquisition, Vol. 1: The data, 373–524. Hills-
dale, NJ: Lawrence Erlbaum.
Fillmore, Charles. 1968. The case for case. In Emmon Bach and Robert
Harms, (eds.), Universals in Linguistic Theory, 1–88. New York: Holt, Rinehart
and Winston.
Friedman, William J. 1978. Development of time concepts in children. In H.
W. Ruse and Lewis P. Lipsitt (eds.), Advances in Child Development and Behav-
ior, Vol. 12, 267–298. New York: Academic Press.
Liang, Lu Yao, Dandan Wu and Hui Li. 2019. Chinese preschoolers’ acquisi-
tion of temporal adverbs indicating past, present, and future: a corpus-
based study. Journal of Child Language 46: 760–784.
Lieven, Elena V. M., Julian M. Pine and Gillian Baldwin. 1997. Lexically-based
learning and early grammatical development. Journal of Child Language 24:
189–219.
MacWhinney, Brian and Catherine Snow. 1985. The child language data
system (CHILDES). Journal of Child Language 1: 271–296.
Mandler, Jean M. 1990. A new perspective on cognitive development in
infancy. American Scientist 78: 236–243.
Meisel, Jürgen M. 1994. Getting FAT: Finiteness, agreement, and tense in
early grammars. In Jürgen M. Meisel (ed.), Bilingual First Language Acquisi-
tion: French and German Development, 89–127. Amsterdam: John Benjamins.
Newcombe, Nora and Janelle Huttenlocher. 2003. Making Space: The Development
of Spatial Representation and Reasoning. Cambridge, MA: MIT Press.

Palmer, Frank R. 2001. Mood and Modality. Cambridge: Cambridge University

Press.
Piaget, Jean. 1954. The Construction of Reality in the Child. New York: Basic
Books.
Piaget, Jean. 1971. The Child’s Conception of Time. New York: Ballantine Books.
Pinker, Steven. 1984. Language Learnability and Language Development. Cam-
bridge, MA: Harvard University Press.
Pinker, Steven. 1989. Review of Slobin, Cross-linguistic evidence for lan-
guage making capacity, and Bowerman, What shapes children’s gram-
mar? Journal of Child Language 16: 456–463.
Radford, Andrew. 1990. Syntactic Theory and the Acquisition of English Syntax.
Oxford: Blackwell.
Rovee-Collier, Carolyn K. 1997. Dissociations in infant memory: Rethinking
the development of implicit and explicit memory. Psychological Review 104:
467–498.
Rowland, Caroline F. and Julian M. Pine. 2000. Subject-auxiliary inversion
errors and wh-question acquisition: ‘what children do know?’ Journal of
Child Language 27: 157–182.
Rowland, Caroline F. and Julian M. Pine. 2003. The development of inversion
in wh-questions: A reply to Van Valin. Journal of Child Language 30: 197–212.
Sano, Tetsuya and Nina Hyams. 1994. Agreement, finiteness, and the devel-
opment of null agreements. Proceedings of NELS 24: 543–558.
Shirai, Yasuhiro. 1993. Inherent aspect and the acquisition of tense-aspect
morphology in Japanese. In Heizo Nakajima and Yukio Otsu (eds.), Argu-
ment Structure: Its Syntax and Acquisition, 185–211. Tokyo: Kaitakusha.
Shirai, Yasuhiro. 1998. The emergence of tense-aspect morphology in Japan-
ese: Universal predisposition. First Language 18: 281–310.
Shirai, Yasuhiro and Roger W. Andersen. 1995. The acquisition of tense/
aspect morphology: A prototype account. Language 71: 743–762.
Slobin, Dan I. 1985. Crosslinguistic evidence for the language-making cap-
acity. In Dan I. Slobin (ed.), The Crosslinguistic Study of Language Acquisition,
Vol. 2, 1175–1256. Hillsdale, NJ: Lawrence Erlbaum.
Slobin, Dan I. 2001. Form–function relations: How do children find out what
they are? In Melissa Bowerman and Stephen C. Levinson (eds.), Language
Acquisition and Conceptual Development, 406–449. Cambridge: Cambridge
University Press.
Smoczyńska, Magdalena. 1985. The acquisition of Polish. In Dan I. Slobin
(ed.), The Cross-Linguistic Study of Language Acquisition, Vol. 1, 595–686. Hills-
dale, NJ: Lawrence Erlbaum.
Stephany, Ursula. 1986. Modality. In Paul Fletcher and Michael Garman
(eds.), Language Acquisition, 375–400. Cambridge: Cambridge University
Press.
Stoll, Sabine. 1998. The role of lexical Aktionsart for the acquisition of
Russian aspect. First Language 18: 351–378.

Spelke, Elizabeth S. 1991. Physical knowledge in infancy: Reflections on

Piaget’s theory. In Susan Carey and Rochel Gelman (eds.), The Epigenesis of
Mind: Essays on Biology and Cognition, 133–170. Hillsdale, NJ: Lawrence
Erlbaum.
Spelke, Elizabeth S. 2000. Core knowledge. American Psychologist 55:
1233–1243.
Toivainen, Jorma. 1980. Inflectional Affixes Used by Finnish-Speaking Children
Aged 1–3 Years. Helsinki: Suomalaisen Kirjallisuuden Seura.
Tomasello, Michael. 1992. First Verbs: A Case Study of Early Grammatical Develop-
ment. Cambridge: Cambridge University Press.
Tomasello, Michael. 2000. First steps toward a usage-based theory of lan-
guage acquisition. Cognitive Linguistics 11: 61–82.
Tomasello, Michael. 2003. Constructing a Language: A Usage-Based Theory of
Language Acquisition. Cambridge, MA: Harvard University Press.
Van Valin, Robert D., Jr. 1991. Functionalist linguistic theory and language
acquisition. First Language 11: 7–40.
Van Valin, Robert D., Jr. 1992. An overview of ergative phenomena and their
implications for language acquisition. In Dan I. Slobin (ed.), The Cross-
linguistic Study of Language Acquisition, Vol. 3, 15–38. Hillsdale, NJ: Lawrence
Erlbaum.
Robert Van Valin (ed.), Advances in Role and Reference Grammar, 1–164.
mechanisms of language acquisition. In Michael Tomasello (ed.), The New
Psychology of Language: Cognitive and Functional Approaches to Language Struc-
ture, 221–249. Mahwah, NJ: Lawrence Erlbaum.
guage 29: 161–175.
Van Valin, Robert D., Jr. 2005. The Syntax–Semantics–Pragmatics Interface: An
Introduction to Role and Reference Grammar. Cambridge: Cambridge Univer-
sity Press.
Press.
Weist, Richard M. 1986. Tense and aspect. In Paul Fletcher and Michael
Garman (eds.), Language Acquisition, 356–374. Cambridge: Cambridge Uni-
versity Press.
Weist, Richard M. 1990. Neutralization and the concept of subject in child
Polish. Linguistics 11: 1332–1348.
Weist, Richard, M. 2002. The first language acquisition of tense and aspect.
In Rafael Salaberry and Yasuhiro Shirai (eds.), The L2 Acquisition of Tense-
Aspect Morphology, 25–78. Amsterdam: John Benjamins.

Weist, Richard, M. 2009. One-to-one mapping of temporal and spatial rela-

tions. In J. Guo, E. Lieven, N. Budwig, S. Ervin-Tripp, K. Nakamura and S.
Özçalişkan (eds.), Crosslinguistic Approaches to the Psychology of Language:
Research in the Tradition of Dan Isaac Slobin, 69–80. Hillsdale, NJ: Lawrence
Erlbaum.
Weist, Richard M. 2014a. Developing temporal systems. Psychology of Lan-
guage and Communication 18(2): 126–142.
Weist, Richard M. 2014b. Future temporal reference in child language. In
Philippe De Brabanter, Mikhail Kissine and Saghie Sharifzadeh (eds.),
Future Times: Future Tenses, 87–113. Oxford: Oxford University Press.
Weist, Richard M. and Ewa Buczowska. 1987. The emergence of temporal
adverbs in child Polish. First Language: 7, 217–229.
Weist, Richard M., Aleksandra Pawlak and Jenell Carapella. 2004. Syntactic–
semantic interface in the acquisition of verb morphology. Journal of Child
Weist, Richard M., Aleksandra Pawlak and Karen Hoffman. 2009. Finiteness
systems and lexical aspect in child Polish and English. Linguistics 47(6):
1321–1350.
Weist, Richard M., Hanna Wysocka, Katarzyna Witkowska-Stadnik, Ewa
Buczowska and Emilia Konieczna. 1984. The defective tense hypothesis:
On the emergence of tense and aspect in child Polish. Journal of Child
Language 11: 347–374.
Notes
1 Influenced by Braine’s (1976) ideas regarding positional learning, Lieven

et al. (1997) found further evidence for the argument that children learn-
ing English base their early utterances on ‘low-scope, idiosyncratic formu-
lae’ (p. 189). However, admittedly, the authors were unable to see how the
knowledge of these formulae motivated the child’s innovative develop-
ment into morphosyntactic information processing. Ironically, Braine
(1992) also proposed an acquisition model having the fundamental con-
cepts of RRG with the potential to understand further language develop-
ment (see Van Valin and LaPolla 1997: 643).
2 Regarding cognitive complexity in development, Weist and Stephany held
opposing views. Like Antinucci and Miller (1976) and others, Stephany
accepted the Piagetian theory of cognitive development which imagined
serious limitations in the child’s thought processes. In particular, the
potential for conceptual time travel into the past and the future was
theoretically unavailable, making tense in child language deficient,
lacking deictic relations. Weist et al. (1984) and elsewhere (for example,
Weist 2014b) made the opposite argument.
3 Liang et al. (2019) investigated the acquisition of temporal adverbs in
Mandarin. The youngest group of children in their study was 2;6. While

their repertoire was limited, the 2-year-old children used temporal

adverbs to locate events in the past, present and future. Regarding the
layered structure of the clause (see 18.1), this constitutes evidence for core
layer modification in a tenseless language.
4 Palmer (2001: 13) classifies desire and intention outside the deontic modal
domain. Choi (1995: 179, Table 2) included a comparison of the early
acquisition of sentence-ending suffixes with evidential meanings (e.g. -ta
and -e) with the later acquisition of modal auxiliaries having deontic
meanings (e.g. -ya tway (obligation) and -su isse (ability)).
5 Concerning the acquisition of tense and aspect, Weist (2002) reviewed
mismatches between the acquisition predictions derived from the Prin-
ciples and Parameter framework and the child language data, for
example, the ‘principle of economy of derivation’, Sano and Hyams
(1994: 551). In contrast, Weist (2009) showed how Slobin’s (1985, 2001)
information processing theory predicts ‘typological bootstrapping’ within
a comparison of the acquisition of temporal and spatial systems in Polish
and Finnish children.

19
Grammatical Aspects
of Language Processing
in the Brain
A Role and Reference Grammar Perspective
Abbreviations

∴ therefore PP prepositional phrase

A actor PPT past participle
aSTG anterior superior temporal PrCS pre-core clot
gyrus
BA Brodmann’s area PrDP pre-detached position
CLM clause-linkage marker PRED predicator
IF illocutionary force PSA privileged syntactic
argument
INGR ingressive RH right hemisphere
LH left hemisphere RP reference phrase
LS logical structure RRG Role and Reference Grammar
LTI linking template inventory TNS tense
MR macrorole U undergoer
NMR non-macrorole V verb
NP noun phrase VP verb phrase
NUC nucleus XP unspecified phrase type
19.1 Introduction
As noted in Chapter 1, Role and Reference Grammar (RRG) strives to be a

component of a model of the communicative competence of a native
speaker of a human language, and following Kaplan and Bresnan (1982),
it is incumbent upon theories making such a claim to be implementable in

testable models, psycholinguistic or computational, of language process-

ing.* Nolan (Chapter 21) illustrates various computational implementa-
tions of RRG. Van Valin (2006) attempted to make the RRG linking system
compatible with the results of psycholinguistic investigations of sentence
processing. With respect to production, the RRG linking system was found
to fit well with a psycholinguistic model of the speaker. With respect to
comprehension, the key idea was the precompiling of aspects of the RRG
linking system in semantic and syntactic representations in order to facili-
tate it. This chapter will expand the discussion into the realm of cognitive
neuroscience, specifically the organization of the language processing
system in the brain, and the analysis to be presented builds on the
2006 proposal.
The discussion begins with comprehension and is oriented around
offering a solution to a long-standing mystery about language and the
brain. It concerns language in the right hemisphere of split-brain patients.
This phenomenon raises fundamental questions about the way language is
represented and processed in the brain. The mystery is presented in Section
19.2, followed in Section 19.3 by an overview of ‘dual-stream’ models of
language processing. These models lead to an important question concern-
ing Brodmann’s area 22 in the left temporal lobe, the answer to which
involves the proposals put forth in Van Valin (2006), which is the topic of
Section 19.4. The following section presents the RRG account of the rele-
vant right-hemisphere linguistic abilities of split-brain patients and its
implications for the analysis of certain types of aphasia. The focus in
Section 19.6 then turns to language production, and there is a brief discus-
sion of the implications of the RRG processing model developed for com-
prehension for production. The main points are summarized in the
final section.
19.2 Language in Split-Brain Patients
As a last-ditch effort to control serious epileptic seizures, in the 1940s

surgeons began to sever the corpus callosum, the thick band of fibres
which links the two hemispheres of the brain and is the conduit for infor-
mation travelling between the hemispheres; this procedure is called ‘com-
missurotomy’. This isolated the hemispheres from each other, cutting off
the flow of information between them. Starting in the 1960s, researchers
began the detailed investigation of the cognitive and behavioural conse-
quences of this surgery.1 One of the obvious areas of interest was language,
*
I would like to thank Delia Bentley, Evelina Fedorenko, David Kemmerer, Jean-Pierre Koenig and Anja Latrouite for helpful
discussion and comments on earlier drafts. Some of this research was supported by the German Research Foundation (DFG)
through Cooperative Research Center 991 ‘The structure of representations in language, cognition and science’ at the
Heinrich Heine University Düsseldorf.

Grammatical Aspects of Language Processing in the Brain 695
since the neural structures underlying the grammatical aspects of it are

known to be located in the left hemisphere, and the standard view was
that the primary areas were Broca’s area in the left inferior frontal lobe
(syntax) and Wernicke’s area in the left temporal lobe (semantics) (see
Figure 19.1). It was expected that the isolated right hemisphere would
not have any linguistic capabilities, and beyond single word recognition,
none were found initially (Gazzaniga and Hillyard 1971). ‘In general, what
was meant by “language capacity” in these patients was the ability to
understand written or spoken words’(Gazzaniga 1983: 526). This ability to
understand individual words showed that lexical storage was bilateral,
with each hemisphere potentially having a lexicon of some kind (see also
Zaidel 1983). It should be noted that not all split-brain patients showed this
kind of right-hemisphere linguistic ability.
In the 1980s, Gazzaniga and his colleagues began to test split-brain
patients with the same tasks that were used in the study of what was then
called ‘agrammatic aphasia’. The motivation for this is laid out in Baynes
and Gazzaniga (1988: 120):
Schwartz, Saffran, and Marin (1980) used a two-choice-picture-pointing

paradigm to test the ability of agrammatic aphasics to understand
semantically reversible active and passive sentences (e.g. the boy hit the
girl; the boy was hit by the girl). Agrammatic aphasics were unable to
choose correctly the picture that matched the sentence they heard.
In other words, they could not use the syntactic constraints in the
sentences to determine who hit whom. In contrast, Linebarger,
Schwartz, and Saffran (1983) found that aphasics who were unable to
interpret correct active and passive sentences were able to judge
accurately whether sentences were grammatical.
The Linebarger et al. result is important because it suggests that even
though a patient may not be able to use a grammatical constraint in a
comprehension task, he or she may still know something about
grammar and be able to use this knowledge in other tasks, such as a
grammaticality judgment task.
Baynes and Gazzaniga (1988: 124) report that two split-brain patients show
the same linguistic ability out of their right hemispheres that the ‘agram-
matic’ aphasics in Linebarger et al. (1983) exhibited: ‘The most striking
finding in these two commissurotomized patients with right hemisphere
language is the dissociation between the [impaired-RVV] ability to compre-
hend syntactically constrained sentences and the [intact-RVV] ability to
judge their grammaticality. This dissociation can be found in two distinct
populations, agrammatic aphasics and commissurotomy patients.’ These
results will be discussed in detail in Section 19.5, and an explanation for
them will be proposed.
How, then, is it possible for an isolated right hemisphere to make gram-
maticality judgements? If the rules, representations, constraints, etc., that

constitute a speaker’s grammatical competence are located in the left hemi-

sphere in right-handers, then it should not be possible for right-handed split-
brain patients to show any grammatical competence in their right hemi-
sphere, yet some do. This is the primary phenomenon to be explained.
The first step in answering this question is to review some of the contem-
porary models of language in the brain.
19.3 Models of Language Processing in Contemporary

Cognitive Neuroscience
Until the early 1990s the view that the primary language areas of the brain
are Broca’s and Wernicke’s areas, which are in the left frontal and temporal
lobes, respectively, and which are connected by the arcuate fasciculus, was
widely assumed to be basically correct. Broca’s area includes the pars oper-
cularis and the pars triangularis in the inferior frontal gyrus, while Wer-
nicke’s area involves the superior temporal gyrus and the angular gyrus. The
locations of these two areas are represented in Figure 19.1.
However, evidence that other parts of the brain were directly involved in
language processing began to accumulate. Mazoyer et al. (1993) showed that
when French speakers were asked to listen to well-formed French sentences,
a list of words, and Urdu sentences, there was activation in the anterior
temporal lobe only when French sentences were heard, not in any other
condition. Of particular relevance for this discussion is the claim put forth
in Dronkers et al. (1994) that the anterior part of Brodmann’s area [BA] 222 in
the left temporal lobe is involved in syntactic processing based on a study of
Figure 19.1 Language-related areas of the left hemisphere

Basic drawing of the labelled brain from: A. Blackburn and C. Hwozdek ‘Labeled Brain’ 2016, openclipart.org, licensed
under CCO 1.0: creativecommons.org/publicdomain/zero/1.0/legalcode.

a large number of aphasics of various types. This was an extremely contro-

versial claim back then, in part because it was incompatible with the
traditional model. By the time the full study was published (Dronkers
et al. 2004), it was no longer so controversial, as other studies (e.g. Stroms-
wold et al. (1996), Stowe et al. (1998), Friederici et al. (2000), Friederici et al.
(2003)) found evidence that the anterior part of the left temporal lobe is
involved in morphosyntactic processing.
Recognition of the role of the anterior temporal lobe in syntactic
comprehension is one of the factors that led to the development of what
are called ‘dual-stream models’ of language processing. There are several
(e.g. Hickok and Poeppel (2004, 2007), Saur et al. (2008), Friederici (2009,
2012), Bornkessel-Schlesewsky and Schlesewsky (2013) and Blank et al.
(2016)), and they are by no means in agreement on all points. But they
all agree in positing dorsal and ventral processing streams. The dorsal
stream involves the same brain regions as in the traditional model, while
the ventral stream goes through anterior BA 22 to connect with Broca’s
area (BA 44, 45). In these models, ‘a ventral stream processes speech
signals for comprehension, and a dorsal stream maps acoustic speech
signals to frontal lobe articulatory networks’ (Hickok and Poeppel 2007:
393). The differences among the models will not be explored here; what is
crucial for this discussion is that all of the models mentioned above
include the anterior temporal lobe in the ventral stream.3 Note that BA
22 is not just adjacent to the primary auditory cortex (BA 41, 42) but
surrounds it, a convenient location for a brain region crucially involved
in the comprehension process.
What is the exact role that anterior BA 22 plays in the comprehension
process? The evidence from Dronkers et al. (1994, 2004) and the studies cited
earlier, along with Turken and Dronkers (2011) and a more recent one
involving Icelandic aphasics (Magnusdottir et al. 2013) and magneto-
encephalography (MEG) studies of language comprehension (Herrmann
et al. 2011; Brennan et al. 2012), is that anterior BA 22 plays a role in the
early phase of morphosyntactic processing which is typically referred to as
‘structure building’. Friederici summarizes this as follows:
Based on neurophysiological data, these phrase structure building

processes have been localized in the anterior superior temporal cortex
approximately 120–150 ms after word category information is available
as an early automatic syntactic process (Herrmann et al. 2011, Friederici
et al. 2000, Shtyrov et al. 2003). The involvement of the aSTG [anterior
superior temporal gyrus] during syntactic phrase structure building has
been confirmed by functional MRI (fMRI) studies using syntactic
violation paradigms (Friederici et al. 2003), as well as a natural language
listening paradigm (Brennan et al. 2012). It has been proposed that, in
the adult brain, these processes can be fast because templates of
different phrase structures (e.g. determiner phrase, prepositional

phrase) represented in the aSTG/STS are available automatically once

the phrasal head (e.g. determiner, preposition) is encountered
(Bornkessel and Schlesewsky 2006).
(Friederici 2012: 263)
The notion of ‘structure building’ can be characterized in a number of ways.

One way involves rules of the kind proposed in generative grammar (e.g.
Merge) whereby syntactic structure is constructed. Another way is through
the combination of stored pieces or ‘chunks’ of structure, along the lines of
Tree Adjoining Grammar or constructional approaches.4 It will be argued in
Section 19.4 that RRG makes a very specific claim about what structure is
and how it comes into play in the comprehension process. Thus, anterior BA
22 seems to be involved in what Friederici terms ‘syntactic structure build-
ing’, in a sense to be determined, during the very early phase of language
comprehension and therefore plays a central role in the process. It will turn
out that this ‘structure building’ is not purely syntactic.
What is the function of posterior BA 22, which, along with part of the
angular gyrus, constitutes what is traditionally known as Wernicke’s area? It
is associated with semantic processing and the comprehension of meaning,
both lexical and propositional. ‘Semantic processes at the sentential level
are more difficult to localize. They seem to involve the anterior temporal
lobe, as well as the posterior temporal cortex and angular gyrus (Lau et al.
2008, Obleser et al. 2007)’ (Friederici 2012: 263). The interpretation of sen-
tences involves both anterior and posterior BA 22, and this means that
anterior BA 22 is not restricted to syntactic ‘structure building’ but rather
is central to the comprehension process in multiple ways.
Since the anterior temporal lobe appears to be involved in processing

syntactic information and semantic information at least at the
sentential level, the function of this neuroanatomical region has been
discussed as reflecting general combinatorial processes which are
involved in phrase structure building as well as in semantic
combinatorics (Hickok and Poeppel 2007, Saur et al. 2008).
(Friederici 2012: 263)
BA 22 is multifunctional: the anterior part is claimed to ‘build’ syntax and

together with the posterior part is involved in the interpretation of sen-
tences, and this is all carried out by cortex with the same cytoarchitectonic
properties. This is a profound fact: the same cortical structure is central to
the processing of both syntax and semantics,5 and this suggests strongly that
syntax and semantics are neither as distinct as many theorists assume, nor
are they handled by completely separate neural systems. Indeed, Federenko
et al. (2020: 19) argue: ‘Taking all the available data into consideration, it
therefore seems that a cognitive architecture whereby syntactic processing is
not separable from the processing of individual word meanings is most likely’
[emphasis in original].

This leads to a fundamental question: What does BA 22 process, such that

it appears to have syntactic properties which the anterior part handles and
semantic properties which both parts handle? The answer to this question
requires recourse to the proposals put forth in Van Valin (2006), and it leads
to the answer to the main question as well.
19.4 RRG as a Processing Model
The RRG linking system maps a semantic representation into a syntactic

representation and vice versa, and this is motivated by the fact that speakers
go from meaning to form to utterance as part of speech production, while
listeners go from utterance to form to meaning as part of the comprehen-
sion process. But RRG is not a processing model; it is a model of grammar
which has facets that are incompatible with the results of psycholinguistic
investigations of sentence processing. For example, the syntax-to-semantics
linking algorithm assumes that the parser outputs a complete labelled tree
diagram, which the linking algorithm then maps into the semantic repre-
sentation. However, all of the results from studies of sentence processing
show that interpretation is incremental, with comprehenders starting the
interpretive process at the very first word encountered (Schlesewsky and
Bornkessel 2004). Moreover, it assumes that displaced wh-expressions are
linked last after all of the other arguments have been linked, whereas
studies show that speakers try to resolve the long-distance dependency as
soon as possible (Stowe 1985; Clifton and Frazier 1989; Boland et al. 1995;
Traxler and Pickering 1996; Koenig et al. 2003). These abstractions, while
appropriate for a model of grammar, must be abandoned in a
processing model.
There is a further factor to be taken into consideration, namely, the speed
of processing. In the previous section it was noted that syntactic structure
building starts as little as 120 ms after word recognition, and the subjective
impression that speakers have is that interpretation is normally instantan-
eous. In other words, it seems to language users that they access the
meaning directly.
Because interpretation occurs simultaneously with parsing, the RRG
linking system must be integrated into the parser. One way this could be
accomplished is suggested by the approach to sentence comprehension
proposed in Townsend and Bever (2001). They postulate a two-phase compre-
hension process: the first is termed ‘pseudosyntax’, which is statistical and
yields an initial assignment of syntactic structure and thematic relations,
and the second, which they call ‘real syntax’, is a minimalist derivation to
check the results of the first phase. They adduce a large amount of convin-
cing evidence for the initial phase (2001: 188–209), and the focus here will be
on ‘pseudosyntax’, as it combines parsing and determining the initial inter-
pretation of a sentence. They characterize it as follows.

Pseudosyntax consists of the immediate initial processes that isolate

major phrases, differentiate lexical categories, and assign initial
thematic relations. Pseudosyntax involves recognition of function
morphemes and lexical categories, which segregate and distinguish
phrases and verbs. Assignment of words to syntactic categories and
major phrases coincides with the application of frequent sentence
patterns that assign these phrases to thematic roles. The sentence
patterns that are appropriate for a particular sentence depend on
subcategorization properties of verbs.
(Townsend and Bever 2001: 187)
The ‘frequent sentence patterns’ referred to above are labelled ‘canonical

sentence templates’, with the most frequent one being ‘NVN actor action
patient’. They marshal a substantial amount of experimental evidence in
support of the role of canonical sentence templates in ‘pseudosyntax’ (2001:
Ch 7). Hence, ‘pseudosyntax’ is in essence statistically driven templatic
parsing, in which the templates contain information about the thematic
relations of the XPs. It results in what is called ‘good enough’ comprehen-
sion (Ferreira and Patson 2007).
RRG makes use of syntactic templates, which were introduced in Chap-
ter 1. The core templates are for the most part bare pieces of the layered
structure of the clause which are unspecified with respect to the types of
phrases that fill the slots in the template. Consequently, they also have no
information regarding the semantic function of the possible phrases that
can occur in them. All of that information is supplied by the linking
algorithm. The first step in transforming bare syntactic templates into
‘pseudosyntax’ parsing templates (or, more accurately, ‘linking templates’)
is to augment them with phrasal category information, and accordingly
Townsend and Bever’s default canonical sentence template for English
would be RP-NUC-RP. The enhanced RRG templates reflect the fact that, in
RRG, syntax is exocentric and the major syntactic categories are not projec-
tions of lexical categories. So instead of NP, a projection of the lexical
category noun, there is RP, reference phrase, the category of referring
expressions which may or may not be headed by a noun, and instead of VP
there is NUC, nucleus, the unit of the layered structure of the clause which
houses the predicating expression, which may or may not be verbal in
nature. Thus lexical category information, which seems so important for
the processing of Indo-European languages, is largely irrelevant for lan-
guages in other parts of the world (e.g. the Philippines, North America
(indigenous languages)). What is relevant to parsing and interpretation in
all languages is the function of the word as an argument, predicating
element, or modifier.
The other crucial aspect of canonical sentence templates is the assignment
of thematic relations to phrases in the template. Here again, RRG takes a
different tack from Townsend and Bever. Rather than augmenting the

templates with thematic relations, the RRG templates would have the
semantic macroroles actor (A) and undergoer (U). Hence the default canon-
ical sentence template for English would be RP:A-NUC-RP:U. The advantage
of using macroroles can be seen in (1).
(1) a. Max broke Mary’s chair. Agent-Patient

a0 . Mary’s chair was broken by Max.
b. Max likes Mary. Experiencer-Stimulus
b0 . Mary is liked by Max
c. Max received Mary’s letter. Recipient-Theme
c0 . Mary’s letter was received by Max.
If the templates were enhanced with thematic relations, each of these

sentences would require a separate template in both the active and
passive voices, for a total of six, because the thematic relations are
different in each pair of arguments. In terms of macroroles, on the other
hand, only one template is needed in each voice, because in both active
and passive versions Max is the actor and Mary’s chair/Mary/Mary’s letter is
the undergoer. Using macroroles rather than thematic relations reduces
the number of canonical sentence templates needed. The most common
canonical sentence template for English, RP:A-NUC-RP:U, is given in
Figure 19.2.
This template has the core argument RPs annotated with semantic macro-
roles. In addition, the operator projection (lower part) carries empty represen-
tations for tense and illocutionary force, the obligatory operators in a finite
Figure 19.2 Simple English transitive linking template

Figure 19.3 Intransitive template (verbal)
Figure 19.4 Intransitive template (non-verbal)
(main) clause in English; the exact values of these operators are a function of
the clause being processed. There is also an optional place for peripheral
adjuncts; only the core-level periphery is shown here, but there may be periph-
eries modifying the nucleus, core or clause. In Figures 19.3 and 19.4, the two
basic intransitive templates are given. The first is for verbal nuclei, and the

single RP core argument is not annotated, because it can be either actor or

undergoer, for instance Bill [actor] laughed (at the joke) vs. The gravy [undergoer]
spilled (on the tablecloth). The second is for non-verbal nuclei, such as Sally is a
lawyer [NP]/very tall [AdjP]/in the library [PP], all of which require the auxiliary be.6
The next two templates are for three-place predicates in English. The first
(Figure 19.5) is for sentences with two direct core arguments (the macrorole
arguments) and an oblique core argument or argument-adjunct.
(2) a. Mary presented the first-place trophy to Sam. Oblique core argument
a0 . Mary presented Sam with the first-place trophy. Oblique core argument
b. Sam put the first-place trophy in a box /on the shelf. Oblique argument-adjunct
c. Sam gave the box to his assistant. Oblique core argument
The second (Figure 19.6) is for the so-called ‘double-object’ construction;

in RRG terms there are three direct core arguments, an actor, an under-
goer and a non-macrorole argument (NMR). When both macrorole argu-
ments are present in an active-voice English core, the actor is the core-
initial RP and the undergoer is the immediately post-nuclear RP. Accord-
ingly, in Sam gave his assistant the box, for example, his assistant is the
undergoer, not the box. It should be noted that there are templates for RPs
and PPs (see Chapter 1).
The final templates to be introduced in this initial discussion are for
passive-voice constructions, in Figures 19.7 and 19.8.
The template in Figure 19.7 handles transitive verbs, while the one in
Figure 19.8 is for three-argument verbs which have an oblique third core
argument (e.g. the examples in (2)). Passives of the ‘double-object’ construc-
tion would require a template like that in Figure 19.8 except for the
non-macrorole core argument being an RP rather than a PP. The actor of
the passive verb may optionally occur in a PP headed by by in the core-level
periphery, and other adjuncts (e.g. temporal or locative expressions) may
also appear in the peripheries.
These templates are the first part of the solution to the problems out-
lined at the beginning of the section; they identify the phrases and their
functions and assign semantic macroroles to the two primary arguments
in a transitive or ditransitive clause. As should be clear, from an RRG
perspective, there is nothing ‘pseudo’ about pseudosyntax. The second part
is arriving at the correct interpretation of the sentence, and for this we
have to take a look at the lexicon and the nature of lexical entries. In the
RRG theory of grammar, it is claimed that all of the morphosyntactic
properties of a completely regular verb can be derived from its semantic
representation together with the linking algorithm (see e.g. Van Valin and
LaPolla 1997: 154ff; Van Valin 2005: 66). However, this involves applying
rules, principles and constraints, all of which would slow down a process-
ing system. Hence in a processing model it is better to precompile the
relevant information in the lexical entries for verbs, so that when the verb
is recognized, all of this information becomes available immediately, as
proposed in Van Valin (2006).7

Figure 19.5 Template for three-place predicates
Figure 19.6 Template for ‘double-object’ construction
The two things that are most important for the rapid interpretation of a
template is the assignment of semantic macroroles and non-predicative
prepositions. For a simple transitive verb like kill or smash, the logical
structures [LSs] would be as in (3).

Figure 19.7 Passive template for transitive verbs
Figure 19.8 Passive template for three-place predicates

(3) a. General: [do0 (A: x, [affect0 (x, y)])] CAUSE [INGR pred0 (U: y)]
b. kill: [do0 (A: x, [affect0 (x, y)])] CAUSE [INGR dead0 (U: y)]
c. smash: [do0 (A: x, [affect0 (x, y)])] CAUSE [INGR smashed0 (U: y)]
Once the template in Figure 19.2 has been matched and the verb recognized,
the basic ‘who did what to whom’ meaning is available by integrating the
arguments from the template with the LS. This can be illustrated with The
burglar smashed the window.
1. The matching mechanism selects the template in Figure 19.2, yielding

[CORE [RP A: the burglar] [NUC smashed ][RP U: the window]] (simplified).
2. The lexical entry for the predicate in the nucleus is activated, and the
annotations on the arguments in the template and the LS guide the
integration (i.e. RP: A the burglar ¼ A: x, ∴ the burglar ¼ x, etc.),8 yielding
the interpretation [do0 (A: the burglar, [affect0 (. . .)])] CAUSE [INGR smashed0
(U: the window)].
The assignment of syntactic structure and the semantic interpretation are

accomplished rapidly thanks to the annotations shared by the templates
and the lexical representations. This is accomplished in BA 22 using both
anterior and posterior resources (the ‘sentential combinatorics’ mentioned
in the previous section) as well as other regions in the left temporal lobe.
What was called ‘early structure building’ in the previous section is, on this
view, assembling syntactic templates, that is, selecting the RP templates for
the RP slots in the clausal template, and matching the resulting template
with the utterance being processed.
Things are a bit more interesting when there are more than two argu-
ments. A good example of a three-place verb is the transfer predicate
present; its LS would be [do0 (x, [affect0 (x, z)])] CAUSE [INGR have0 (y, z)].
Present, like most three-place verbs in English, permits more than one
possible choice of the undergoer macrorole, as shown in (2a, a0 ), and this
choice is signalled by the preposition assigned to the non-macrorole argu-
ment. Rather than invoking macrorole assignment rules and preposition
assignment rules as in an RRG grammar, the lexical entry for present
specifies both possibilities, as in (4).
(4) present: a. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: to y, U: z)]
b. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (U: y, NMR: with z)]
Once the template in Figure 19.5 has been matched and the verb recognized,
the basic ‘who did what to whom’ meaning is available by integrating the
arguments from the template with the LS. For verbs like give that occur in
the ‘double-object’ construction (e.g. Sam gave his assistant the box), their LSs
would be as in (5).
(5) give: a. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: to y, U: z)]
b. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (U: y, NMR: z)]

The LS in (5a) integrates with the template in Figure 19.5, while the one in
(5b) integrates with the template in Figure 19.6.
The comprehension of a passive sentence such as The window was smashed
by the burglar is just as straightforward.

[CORE [RP U: the window][NUC was smashed ]][PERIPHERY [PP A: by the burglar]]
(simplified).
2. The lexical entry for the predicate in the nucleus is activated, and the
integration (i.e. PP: A: by the burglar ¼ A: x, ∴ the burglar ¼ x, etc.), yielding
the interpretation [do0 (A: the burglar, [affect0 (. . .)])] CAUSE [INGR smashed0
(U: the window)]
In the case of three-place predicates (e.g. present), it is the coding of the

NMR which is the key to determining which LS is the correct one.
(6) a. The first-place trophy was presented to Sam by Mary.

1. The matching mechanism selects the template in Figure 19.8 yielding
[CORE [RP U: the trophy ][NUC was presented ] [PP NMR: to Sam]][PERIPHERY [PP A:
by Mary]].
2. The lexical entry for the predicate in the nucleus in (4a) is selected
because of the NMR: to specification and is activated, and the
integration (i.e. PP: A: by Mary ¼ A: x, ∴ Mary ¼ x, etc.), yielding the
interpretation
[do0 (A: Mary, [affect0 (. . .)])] CAUSE [INGR have0 (NMR: Sam, U: the trophy)].
b. Sam was presented with the first-place trophy by Mary.
[CORE [RP U: Sam ][NUC was presented ] [PP NMR: with the trophy ]][PERIPHERY [PP
A: by Mary]].
2. The lexical entry for the predicate in the nucleus in (4b) is selected
because of the NMR: with specification and is activated, and the
integration (i.e. PP: NMR: with the trophy ¼ NMR: z, ∴ the trophy ¼ z, etc.),
yielding the interpretation
[do0 (A: Mary, [affect0 (. . .)])] CAUSE [INGR have0 (U: Sam, NMR: the trophy)]
These examples illustrate how the combination of semantic-macrorole-

augmented syntactic templates and annotated lexical entries for verbs can
rapidly yield the basic interpretation of a sentence (i.e. ‘who did what to
whom’). It has been argued (McKoon and Love 2011) that the internal
semantic complexity of a verb affects the speed at which it is processed,
and this is readily captured in the LS representations.
There were two issues mentioned at the beginning of this section that RRG
as a processing model must address, namely incremental interpretation and
displaced wh-questions. Incremental interpretation can be accounted for in

Figure 19.9 English subject wh-Q template
terms of parallel processing of statistically weighted templates in conjunc-

tion with a beam-search algorithm of the type presented in Jurafsky (1996),
which ranks candidate structures and lexical items within a specific range
of probability, dropping candidates that fall outside that range as the
process moves forward. At the first word in a sentence there are many
possible candidate templates, and as each new word is encountered, the
number of candidates is reduced until the correct combination of templates
is chosen.
Displaced wh-questions are handled by the templates in Figures 19.9 and
19.10. A few remarks are in order about these two templates. First, optional
peripheries have been omitted. Second, there are no traces or other kinds of
phonologically null elements in the trees. Third, the subject wh-Q template
is compatible with transitive verbs and three-argument verbs with an
oblique third core argument, while the non-subject wh-Q template works
for verbs with two or more arguments. Intransitive verbs require wh-
Q versions of the templates in Figures 19.3 and 19.4. Fourth, when the verb
is transitive, these templates are active voice; passive voice requires distinct
wh-Q templates. Fifth, these apply only to simple sentences; more complex
templates would be required for long-distance wh-questions (see
Figure 19.11)). An example of a simple wh-question and its analysis is given
in (7). (‘~A’ means ‘not the A argument of the active voice transitive or three-
place verb in the main clause’.)
(7) What did Mary give to Sam?

[PrCS [RP ~A: what ]][CORE {did} [RP A: Mary] [NUC give ] [PP NMR: to Sam]].9

Figure 19.10 English non-subject wh-Q template
Figure 19.11 Long-distance wh-question linking template
2. The lexical entry for the predicate in the nucleus in (5a) is selected because
of the NMR: to specification and is activated, and the annotations on the
arguments in the template and the LS guide the integration (i.e. RP: A:
Mary ¼ A: x, ∴ Mary ¼ x, etc.), yielding the interpretation
[do0 (A: Mary, [affect0 (. . .)])] CAUSE [INGR have0 (NMR: Sam, U: what)].

The ~A wh-expression what is integrated into the LS as the U, since undergoer

is a kind of non-actor.
An example of a long-distance question template is presented in
Figure 19.11. The application of this template to a wh-question is presented
in (8).10
(8) What did Mary say that Sally gave to Sam?

[Clause [PrCS [RP ~A: what ]][CORE {did} [RP A: Mary] [NUC say ]]
[CLM that ] ! [CLAUSE: U [CORE [RP A: Sally] [NUC gave ] [PP NMR: to Sam]]]].
2. The lexical entries for the predicates in the main clause nucleus, say, and in
the subordinate clause nucleus, give, are activated, and the LS form give in
(5a) is selected because of the NMR: to specification, and the annotations on
the arguments in the template and the LSs guide the integration, yielding
the interpretation do0 (A: Mary, [say0 (Mary, U: [[do0 (A: Sally, [affect0 (. . .)])]
CAUSE [INGR have0 (NMR: Sam, U: what)]].
This template works for a variety of verbs but not all (see Shimojo, Chap-
ter 16): What did Mary think/believe/expect/anticipate/claim/deny/doubt/assert that
Sally gave to Sam? These templates directly address the interpretation of the
wh-expression and do not put it off until the rest of the sentence has been
processed, as the linking algorithm does.
The template in Figure 19.11 introduces templates for complex sentences,
and in this case the structure involves subordination (embedding). There are
non-embedded complex sentences as well, and sentences like Mary persuaded
Sam to wash the dishes are represented by templates like the one in
Figure 19.12.
Figure 19.12 Infinitival complement template

Sentences with this structure (e.g. Mary persuaded/promised Sam to wash the
dishes) obligatorily ‘share’ an argument; that is, the infinitival complement
is missing an argument, and accordingly one of the core arguments in the
first core is interpreted as also filling a role in the second core. In this
sentence the actor of wash is missing; when the verb in the first core is
persuade, the undergoer Sam is interpreted as the actor of wash, whereas
when the verb is promise, the actor Mary is so interpreted. The LS for the
persuade version would be [do0 (Mary, [affect0 (Mary, Sam)])] CAUSE [want0
(Sam, [do0 (Sam, [wash0 (Sam, dishes)])])]. Despite multiple occurrences in the
LS, the RP Sam appears only once in the sentence as the undergoer of
persuade. These last two examples barely scratch the surface of the intricacies
of complex sentences, but they illustrate how the system can be applied
to them.
What happens when breakdowns such as garden paths occur? What
happens when an unfamiliar structure is encountered? These are situations
in which step-by-step reanalysis and reinterpretation is required. On top of
the template-based processing discussed earlier, there is also the possibility
of step-by-step processing as well. In this mode, the syntactic representation
of the sentence is constructed, and then the steps of the syntax-to-semantics
linking algorithm are carried out literally, which permits double-checking
of all relevant aspects of the construction.
All of the examples of linking with semantically augmented syntactic
templates have been from English, which has a relatively rigid word order
that is crucial for the interpretation of the core arguments. Can this
approach to language comprehension be applied to languages with various
degrees of word-order flexibility and which rely on case marking rather than
linear order for the interpretation of core arguments? The answer is ‘yes’,
and the application to case-marking languages with very flexible word order
will be illustrated with an example from Croatian.
(9) darovati ‘give as a gift’ [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: y, U: z)]
a. Unuc-i su bak-i darova-l-i cvijeć-e.
grandson-m.pl.nom be.3pl grandmother-f.sg.dat give-pst-pl flower-m.pl.acc
‘The grandsons [x] gave flowers [z] to [their] grandmother [y].’
b. Baki su unci darovali cvijeće.
c. Cvijeće su baki unci darovali.
d. Baki su darovali cvijeće unci.
The Croatian verb darovati ‘give as a gift’ is a three-argument transfer verb

like English give or present. The LS in (9) has the same semantic annotations
as in (4a) and (5a), and it represents the meaning of all of the sentences in
(9a–d) and the many other possible arrangements of these five words. The
only thing that is invariable in terms of word order is the occurrence of the
word su as the second word; it is a kind of auxiliary verb which shows person
and number agreement with the nominative RP and must always be the
second word in the sentence.11

Interestingly, darovati shows the same undergoer alternation that present

does.
(10) a. Unuc-i su bak-u darova-l-i cvijeć-em.

grandson-m.pl.nom be.3pl grandmother-f.sg.acc give-pst-pl flower-m.pl.ins
‘The grandsons [x] gifted [their] grandmother [y] with flowers [z].’
b. Baku su unci darovali cvijećem.
c. Cvijećem su baku unci darovali.
d. Baku su darovali cvijećem unci.
As in (2a0 ) and the LS in (4b), but not in (9), the y-argument in the LS is the
undergoer, and the z-argument is the non-macrorole core argument in the
instrumental case, which is analogous to the instrumental preposition with
in English. As in (9), all of the examples in (10) have the same meaning. The
crucial difference between (9) and (10) is the coding of the non-macrorole
argument, dative in (9) and instrumental in (10), and accordingly it must be
specified in the LSs for darovati, as in (11), just like in (4) for English present.
(11) darovati ‘give as a gift’

a. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: dat y, U: z)] ¼ (9)
b. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (U: y, NMR: INST z)] ¼ (10)
It is clear from the examples in (9) and (10) that linear order plays no role
in the interpretation of these sentences and that the interpretation of the
core arguments is based on case marking. In Van Valin (2006) ‘case associ-
ation principles’ were used to handle the linking between case forms and
macroroles. The grammar of Croatian includes the ones in (12) as well as
additional ones for cases which are not found in (9) and (10).
(12) Croatian case association principles

a. NOM [Actor > Undergoer]
b. ACC [Undergoer]
c. DAT [NMR]
d. INST [NMR]
The nominative case codes the actor with some intransitive verbs (those
with an activity predicate in their LS) and active-voice transitive and ditran-
sitive verbs, and it codes the undergoer with some intransitive predicates
(those without an activity predicate in their LS) and with passive-voice
transitive and ditransitive verbs. The accusative case marks the undergoer
of an active-voice transitive or ditransitive verb. The other two cases mark
non-macrorole arguments, dative arguments being direct core arguments,
while instrumental ones are oblique core arguments.
In RRG, whether an XP is a core argument or a peripheral adjunct is a
consequence of the semantic function of the XP and not on its place in the
linear string. (Languages may impose linearization restrictions, of course.)
This means that the syntactic templates which represent the layered struc-
ture of the clause need not have a fixed order, and this is what languages
like Croatian require: unordered templates. The only ordering constraint in

Figure 19.13 Croatian linking template for (9)
the clause concerns the auxiliary, which must be the second word in all
clauses. This can be represented as in Figures 19.13 and 19.14.
The dotted lines connecting the RPs and the nucleus to the core node
indicate that the order of these elements is not fixed but that any of
them could occupy any of the positions. The solid line connecting the
AUX node to the core node signals, on the other hand, that its position
in the core is fixed and it must invariably be in the second position.
Information structural considerations like the relative topicality vs.
focality of the referents of the RPs strongly influences the order of words
in context, as is well known. Croatian not only allows the RPs and
nucleus to occur in any order, but it also permits discontinuous constitu-
ents, as in (13a) vs. (13b).
(13) a. Naš-a učionic-a je udobna.

our-nom classroom-nom be.3sg comfortable
‘Our classroom is comfortable.’
b. Naš-a je učionic-a udobna.
our-nom be.3sg classroom-nom comfortable
‘Our classroom is comfortable.’ [literally ‘Our is classroom comfortable.’]
Je ‘is’ is like su ‘are’ in (9) and (10), as it must be second in the clause, and
second position may be defined as being after the first constituent, as in
(13a), or after the first word, as in (13b). (Native speakers prefer (13b).)
Determiners are connected to the NUCLEUSR via the operator projection of

Figure 19.14 Croatian linking template for (10)
the RP, and Croatian, unlike English, does not require strict adjacency
among RP constituents, nor is their order fixed. Hence in Croatian linking
templates like those in Figures 19.13 and 19.14, not only are the constituents
in the clause unordered, but there are few constraints on the RP templates
as well.12 These templates apply, accordingly, to all of the word-order vari-
ants in (9) and (10), as well as others not exemplified, and other three-
argument verbs with a dative or instrumental third argument. The match-
ing process goes as follows for (9c), for example.

[CORE [RP ACC:U: Cvijeće] [AUX su ] [RP dat:NMR: baki][RP nom:A: unci] [NUC darovali ]]
2. The lexical entry for the predicate in the nucleus in (11a) is selected
because of the NMR: DAT specification and is activated, and the annota-
tions on the arguments in the template and the LS guide the integration
(i.e. RP:NOM:A: unci ¼ A:x, ∴ unci ¼ x, etc.), yielding the interpretation
[do0 (A: unci, [affect0 (. . .)])] CAUSE [INGR have0 (NMR: baki, U: cvijeće)]
To sum up, the determination of the syntactic structure and the semantic
interpretation of an utterance is achieved rapidly due to the annotations on
both the templates and the lexical representations of predicates. This is
carried out in BA 22 using both anterior and posterior resources, the ‘sen-
tential combinatorics’ mentioned in the previous section, as well as other
areas of the left temporal lobe. What was called ‘early structure building’ in

the previous section is, on this view, the assembling of syntactic templates,
for instance selecting the RP templates for the RP slots in the clausal
template, and the matching of the resulting template with the utterance
being processed. The integration of the information in the lexical represen-
tation of the predicate in the nucleus of the clause yields a basic ‘who did
what to whom’ meaning (a ‘good enough’ interpretation). In the case of
breakdowns, the system can execute the syntax-to-semantics linking algo-
rithm in a controlled, step-by-step fashion.
At the end of Section 19.3, the following question was raised: What does
BA 22 process, such that it appears to have syntactic properties which the
anterior part handles and semantic properties which both parts handle? The
answer should now be clear: semantic-macrorole-augmented syntactic
linking templates and the lexical entries for predicates. Given the integra-
tion of both syntactic and semantic information in the linking templates
and in the lexical entries for predicates, it is in fact difficult to claim that
anterior BA 22 is primarily for syntactic processing and posterior BA 22 is for
semantic processing. Rather, all of BA 22 is involved both aspects of lan-
guage comprehension, consonant with the point made by Federenko et al.
(2020) mentioned earlier. They are also crucial for the resolution of the
mystery stated at the end of Section 19.2: How is it possible for the isolated
right hemisphere in a split-brain patient to make grammaticality
judgements?
19.5 Language in Split-Brain Patients Revisited
The linguistic capacity of the right hemisphere in split-brain patients is not

uniform; as pointed out in Section 19.2, some patients show no linguistic
ability at all, some are able to recognize single words, and a small minority
are capable of making grammaticality judgements. The latter two groups
are both relevant for this discussion, not just the third. The fact that some
patients can understand individual words is evidence, following Gazzaniga
(1983) and Zaidel (1983), that storage of lexical items in the brain is bilateral
in some people. RRG, like many (but not all) theories, postulates that the
grammar of a human language has a lexicon containing (minimally) the
words and morphemes of the language, and the lexical entries for verbs
would contain the LS(s) for the verb (e.g. (3b, c), (4a, b), (11a, b)), including the
precompiled information on macrorole assignment and the preposition or
case of non-macrorole arguments. What the research on commissur-
otomized patients has shown is that in some people the lexicon is in both
hemispheres, not just the left. RRG grammars also have a repository for the
syntactic templates that play a role in the linking from semantics to syntax;
it is called the ‘syntactic inventory’. These are the same syntactic templates
that are augmented with morphosyntactic and semantic macrorole infor-
mation to create the linking templates at the heart of the RRG processing

system developed in the previous section. These linking templates must be

stored somewhere, and so it is appropriate to posit the existence of a ‘linking
template inventory’ (LTI). The lexicon and the LTI house a language’s lexical
and core grammatical resources, respectively, and given that the lexicon is
bilaterally instantiated in some speakers,13 it is reasonable to assume that
the LTI is as well.
The situation prior to commissurotomy is, therefore, as follows for at least
some right-handed native speakers. Lexical resources are stored in both
hemispheres, and the mechanisms for inflecting and deriving words are
located in the left hemisphere. Grammatical resources involved in compre-
hension are of two kinds. First there are the linking templates, which
minimally contain phrase-structural information and the semantic
(macro)roles of the arguments in the structure. They may also contain
syntactic, semantic and/or information-structural constraints on the con-
struction represented by the template. As an example of these constraints,
consider reflexive constructions, as in Bill saw himself in the mirror. They are
subject to a variety of constraints governing the choice of the controller
(antecedent, i.e. Bill), properties of the reflexive pronoun (himself), and how
far away the reflexive pronoun can be from the controller. A rough approxi-
mation of the constraints in English, German and Icelandic is given in (14);
the semantic role hierarchy is, in a generic, oversimplified form, agent >
experiencer > location/goal/source > theme > patient.
(14) Restrictions: Controller Reflexive Syntactic domain

a. English Higher on Agrees with Simple clause
semantic role controller person, (not across more
hierarchy number, gender than one core)
b. Icelandic Highest Special form only Complex clause
ranking direct 3 per. No number (can cross more
core argument or gender than one core)
[MR] agreement [sig]
c. German Older spkrs: Same as Same as English
NOM only Icelandic
Younger: like [sich]
Icelandic
There are both semantic and morphosyntactic constraints here. English is

unique among the three languages in allowing more than one possible
controller in a clause, as in Bill talked to Sam about himself, where either Bill
or Sam can be the one being talked about, something that is unthinkable in
the other two languages with sig/sich. Older German speakers have a unique
morphologically defined controller, the nominative RP, while Icelanders
and younger German speakers have a semantically based constraint. The
latter two languages have an invariable reflexive anaphor in terms of person
(3rd only), number and gender, in contrast to English. Finally, in the struc-
ture represented in Figure 19.12, Bill asked Sally to help himself is possible in
Icelandic but not in German or English. These constraints would be part of

the linking template for that construction, which would share structural
and semantic macrorole properties with non-reflexive constructions (e.g. Bill
asked Sally to help Mary).14 Like lexical items, these templates are stored in
both hemispheres. The second kind of grammatical resource consists of the
mechanisms for selecting and interpreting templates; they are primarily in
the left temporal lobe, including BA 22. The interpretation of templates
includes the integration of the semantic representations of lexical items
with the templates based on the semantic role annotations and other infor-
mation, as well as the evaluation of the sentence with respect to any
constraints on the construction. Thus what may be termed ‘comprehension
competence’ is distributed: it is found in the content of the linking tem-
plates, in the content of the lexical representations of verbs and other
predicates, and in the interpretive mechanisms.
After the commissurotomy, what are the linguistic resources in the right
hemisphere? At most there would be lexical items in the lexicon and linking
templates in the LTI. They are cut off from the left-hemisphere interpretive
mechanisms mentioned earlier. Yet a few commissurotomized patients are
able to make grammaticality judgements using their right hemisphere. The
main grammatical information available in their right hemisphere is in the
linking templates, namely, the structure of phrases and clauses, and there is
minimal grammatical information in the lexical entries for verbs: while the
LSs cannot be interpreted semantically, the number of arguments and the
preposition(s) marking oblique arguments can be read off the LSs. Conse-
quently these two sources of information must be the basis for the gram-
maticality judgements. The issue, then, is how is this grammatical
information sufficient for making grammaticality judgements but insuffi-
cient for comprehending ‘syntactically constrained sentences’ like revers-
ible passives, as discussed in Section 19.2.
The linking templates contain three kinds of information: a representa-
tion of phrase structure, the semantic roles of the arguments in the
structure, and construction-specific constraints. The semantic role infor-
mation and the construction-specific constraints (e.g. (14)) cannot be pro-
cessed in the right hemisphere, and so they are in effect invisible. The two
kinds of information express features overlaid on the phrase structure, and
they require the sophisticated neural mechanisms found in the left tem-
poral lobe for their interpretation. This accounts for the inability to cor-
rectly interpret reversible passives. The default strategy of interpreting the
first argument as the actor is a manifestation of the ‘subject preference’
(Wang et al. 2009), which has been found to be operative across languages,
and not from an ability to interpret some of the semantic role annotations
on the templates but not others. It is primarily the phrase-structure infor-
mation that makes possible the grammaticality judgements through the
mechanism of pattern matching, something the right hemisphere excels
at. It is heavily involved in face recognition, for example, and has its own
pattern-matching resources independent of the left hemisphere, unlike

the situation with language. Hence the hypothesis is that split-brain

patients who exhibit grammatical behaviour out of their right hemisphere
are basing their grammaticality judgements on matching the input sen-
tence with a phrase-structure pattern represented in a linking template.
Confronted with sentences like Are you going to the store? and *Are going you
to the store?, the patient’s right hemisphere can find a template matching
the pattern in the first sentence, but there is no template with the pattern
of the second sentence, and therefore it must be judged as ungrammatical.
It may seem odd that a passive sentence can be judged grammatical but
cannot be interpreted, but this follows from the pattern-matching hypoth-
esis: passive sentences like those in (15a–c) show the same basic pattern
(RP-V-PP) as the intransitive sentences in (15a0 –c0 ), namely the one in
Figure 19.3.
(15) a. Bill was chased by a big dog.

a0 . Bill was running by a big dog.
b. Donald is being interviewed on Fox News.
b0 . Donald is binging on Fox News.
c. Sally has been arrested by the police.
c0 . Sally has been waiting by the police.
The morphological contrast between present and past participles, which is

crucial to the interpretation of these sentences, is not represented as a
structural feature of the templates in question and is, therefore, difficult
to apprehend, apparently. Nevertheless, the passive sentences can be
matched with the RP-V-PP template in Figure 19.3 and therefore judged to
be grammatical.
This account makes a prediction that, if correct, would provide empirical
support for it. Baynes and Gazzaniga (1988) report the results of having their
commissurotomized patients undertake the same ten tests that Linebarger
et al. (1983) gave their ‘agrammatic’ aphasic subjects. The prediction is:
subjects’ judgements will be more accurate when the issue is phrase-
structural in nature and therefore involves templates and pattern matching
than when it involves a non-phrase-structural issue that does not involve the
phrase-structural aspects of templates. The ten tests are given in (16) along
with a specification as to whether the issue is phrase-structural or not,
followed by the scores of the two hemispheres of J.W., one of the patients
discussed in Baynes and Gazzaniga (1988), on an index of sensitivity to
grammaticality, A0 (see Linebarger et al. 1983: 379). On this index, 1.0 is a
perfect score, meaning all examples were correctly identified as grammat-
ical or ungrammatical.
(16) J.W. LH15 J.W. RH

16
a. Strict subcategorization of verbs Phrase-structural .91 .90
b. Particle movement Phrase-structural .92 .97
c. Subject-aux inversion Phrase-structural .90 .80
d. Empty elements17 Phrase-structural .89 .77

e. Tag questions (co-indexing) Non-structural .92 .77

f. Left branch condition18 Phrase-structural .94 .92
g. Gapless relative clauses Phrase-structural .92 .89
h. Phrase structure rules Phrase-structural .75 .90
i. Reflexives (co-indexing) Non-structural .85 .65
j. Tag questions (aux copying) Non-structural .92 .74
Mean accuracy .93 .84
J.W.’s right hemisphere performed surprisingly well, and the three tests
with the biggest difference in performance between the two hemispheres
were the three non-phrase-structural tests involving agreement and/or co-
indexing. Baynes and Gazzaniga (1988) comment on J.W.’s performance:
As Fig. 1 [¼ (16)] illustrates, J.W.’s left hemisphere performs well.

Although less accurate, his right hemisphere has the most difficulty
with the same conditions as the agrammatic aphasics. Inspection of
Fig. 1 also indicates that there is a much tighter relation between the
pattern generated by J.W.’s right hemisphere and that generated by the
aphasics. Moreover, the pattern of his left hemisphere appears to be
distinct from that of his right hemisphere. The failure to find any
interhemispheric similarity suggests that cross-cueing is not accounting
for the good performance of the right hemisphere. (123)
The other patient discussed, V.P., had nearly identical performance in both
hemispheres.
V.P.’s results are more difficult to interpret because of the lack of a

convincing difference between her left and right hemispheres. It is
especially important to establish a difference in this subject because of
the demonstration that some fibers of the splenium remain intact.
Because V.P. has previously demonstrated greater right hemispheric
syntactic competence than J.W. and because she appears to have some
ability to generate speech from her right hemisphere, it would not be
surprising that her right hemisphere would do well at this task.
Without a greater dissociation of right and left hemisphere
performance, little more can be said about V.P.’s performance at
this time.
(Baynes and Gazzaniga 1988: 124)
It seems clear that V.P.’s very similar performance in both hemispheres

sheds little light on the prediction being tested, especially given that her
two hemispheres are not completely isolated from each other, but the
results from J.W.’s right hemisphere strongly suggest that the prediction is
correct, because the three tests on which his right hemisphere performed
most poorly in comparison with his left hemisphere were precisely the three
tests that were not phrase-structural; they involve agreement and corefer-
ence, not phrase structure. Thus, the idea that the grammatical competence
exhibited by the right hemisphere after commissurotomy derives from the

interaction of linking templates together with the pattern-matching abil-

ities of the right hemisphere is plausible and provides an explanation for
J.W.’s (and possibly V.P.’s) grammatical abilities,19 and this in turn supports
the RRG approach to language comprehension laid out in Section 19.4. The
question raised at the end of Section 19.2 has been given a plausible,
theoretically motivated answer.
It might well be objected that this is a rather far-reaching conclusion to be
drawn based on data from only one patient, at most two. But J.W. is in fact not
unique; there are also the four aphasics in Linebarger et al. (1983), who
exhibited the same dissociation as J.W. As mentioned back in Section 19.2,
Baynes and Gazzaniga (1988: 124) comment ‘[t]he most striking finding in
these two commissurotomized patients with right hemisphere language is
the dissociation between the [impaired-RVV] ability to comprehend syntactic-
ally constrained sentences and the [intact-RVV] ability to judge their gram-
maticality. This dissociation can be found in two distinct populations,
agrammatic aphasics and commissurotomy patients.’ They were tested on
the same materials, and ‘[t]he conditions in which Linebarger et al.’s original
four aphasic subjects were consistently weakest were 5[¼(16e)], 9[i], and 10[j].’
(1988: 122–123) Those are precisely the three non-phrase-structural tests and
the three that J.W. performed the worst on. Because of this, they conclude: ‘As
J.W.’s right hemisphere shows a similar pattern of performance it seems likely
that it employs the same mechanism as that of the agrammatic aphasics’
(1988: 124). However, an alternative conclusion would be that the aphasics use
the same mechanism as J.W.’s right hemisphere.
What are the consequences of the alternative conclusions? They turn out
to be of great significance. Linebarger et al. (1983) is cited as a seminal work
which demonstrates that aphasics still retain their grammatical knowledge
but cannot access it or get it to function in the comprehension process. They
can access their grammatical knowledge only enough to make grammatical-
ity judgements.
The results themselves, however, and their implications for the status of
syntactic knowledge in agrammatism, are quite clear: the
comprehension deficit in agrammatism does not reflect loss of the
capacity to analyse syntactic structure. We are left now with the
important question of why agrammatic aphasics seem unable to make
use of this capacity in the comprehension and production of sentences.
(Linebarger et al. 1983: 390)
Linebarger et al. reject a number of alternative explanations for the dissoci-

ation, including template matching: ‘In general, any appeal to “template
matching” as an alternative strategy for performing this task seems to us to
require templates which are in fact generalizations over structures of consid-
erable syntactic specification’ (1983: 386). The RRG linking templates are just
exactly that, as the discussion of them in Section 19.4 made clear. They are
syntactically specific, and well motivated empirically and theoretically.

It should be noted that this dissociation between the capacity to compre-

hend and the ability to make grammaticality judgements has been replicated
by numerous scholars in a wide variety of languages, for example Schwartz
et al. (1987), Wulfeck (1988), Lukatela et al., (1988) for Serbian, Wulfeck et al.
(1991) for Italian, and Lu et al. (2000) for Mandarin Chinese.
There are two glaring problems for Linebarger et al.’s account. First, it
offers no explanation for how these aphasics make grammaticality judge-
ments and why they are limited to this. Second, it does not apply to J.W. His
grammatical knowledge is fully intact in his left hemisphere, and his right
hemisphere does not contain any grammatical knowledge of the kind
assumed by Linebarger et al. Further, J.W.’s performance on the tests in
(16) cannot be attributed to partial, constrained access to his grammatical
competence, which is locked away in the left hemisphere with no way to
access it from the right hemisphere. If one assumes the analysis given by
Linebarger et al., the fact that J.W. performed the same as the aphasics is a
striking coincidence, nothing more. Another striking coincidence is found
in a comparison of Linebarger et al.’s subjects, J.W.’s results and the results
reported in Wulfeck (1988), a replication of the Linebarger et al. study.
(17) Overall mean A0 (sensitivity to grammaticality)

Linebarger et al. (1983: 379) Highest .94 All: .90 Lowest: .84
Wulfeck (1988: 76) Normal controls: .972 Aphasics: .821
Baynes and Gazzaniga (1988: 123) J.W.’s LH: .93 J.W.’s RH: .84
Linebarger et al.’s subjects have mean A0 values of .94, .91, .91 and .84, which
yields an overall value of .90. The best-performing subject is comparable to
J.W.’s left hemisphere and near the normal controls in Wulfeck’s study,
whereas the worst performer is comparable to the aphasics in Wulfeck’s
study and J.W’s right hemisphere. The similarities in the performance of
J.W.’s right hemisphere and that of the aphasics in the two studies argue
against a solution to the aphasic’s abilities which necessarily excludes J.W.
from consideration.
What about the alternative conclusion, namely, that the aphasics use the
same mechanism as J.W.’s right hemisphere? The explanation given for
J.W.’s behaviour applies equally to the aphasics, and it resolves the paradox
inherent in Linebarger et al.’s interpretation of their results, namely, why is
an aphasic’s grammatical competence accessible for making grammaticality
judgements but not other tasks? The answer is that grammatical knowledge
in the form of linking templates is preserved, but the processing mechan-
isms in BA 22 and other parts of the left hemisphere are out of commission.
Consequently, the aphasics rely on the intact right hemisphere with its
lexicon, LTI and pattern-matching capacity to make grammaticality judge-
ments, just like J.W.’s right hemisphere.
This explanation is not limited to fixed-word-order languages like English.
For a language like Croatian, the analysis presented here predicts that, just
as in English, the argument structure information of verbs will be preserved

in the lexical entries in the bilaterally instantiated lexicon, but, unlike in

English, they will be formulated in terms of morphological cases (cf. (11)).
The linking templates (cf. Figures 19.13, 19.14) are inherently unordered
with respect to the positions of the nucleus and the case-marked RP argu-
ments. The grammaticality judgements made by aphasics would involve
pattern matching, just as in English, but the matching would be based on
the case arrays in the sentence and in the linking template. This crucially
depends on sensitivity to morphological case being intact in the aphasics,
and this is what Lukatela et al. (1988) found in their study of Serbian
aphasics, which concludes ‘the findings of the present study are consistent
with other research both on richly inflected languages and on fixed word-
order languages like English . . . A comparison of agrammatics’ performance
across tasks shows that subjects who standardly fail in an object-
manipulation task may succeed in a grammaticality judgment task that
tests comprehension of the same linguistic structures’ (1988: 13). Thus, the
account developed here can explain the Serbian results in terms of the same
mechanism that was given for the English results, namely pattern matching
with linking templates.
19.6 Implications for Semantics-to-Syntax Linking

and Language Production
This chapter has up to now been concerned with language comprehension,

which is modelled in an RRG grammar by the syntax-to-semantics linking
algorithm. In order to create a plausible processing system, two major
changes to the system were proposed. First, the bare syntactic templates of
the grammar were augmented with syntactic category information (RP, PP)
and semantic macroroles, yielding linking templates. Second, the LSs of
verbs and other predicators were augmented with semantic macroroles
and the adposition or case carried by oblique core arguments. The result
of this is that parsing and syntax-to-semantics linking are collapsed into two
steps, as illustrated for a variety of examples in Section 19.4. What implica-
tions do these changes have for the linking algorithm for mapping seman-
tics into syntax, which underlies possible models of language production?
This question was left open at the end of Van Valin (2006), where it was
noted that there were now two sets of syntactic templates in the theory, the
bare templates used in linking from semantics to syntax, and the semantic-
ally augmented linking templates used in linking from syntax to semantics.
There are two sets of LSs as well: the ‘pristine’ LSs used in semantics-to-
syntax linking, and the morphosyntactically and macrorole-enhanced LSs
used with the linking templates in syntax-to-semantics mapping. There is
clearly a great deal of redundancy here that needs to be resolved.
The semantics-to-syntax linking algorithm was presented in Chapter 1 and
will just be summarized here, using the example Max was interviewed by CNN
during the insurrection. (The operator projection is omitted.)

(18) a. Step 1: Construct the semantic representation of the sentence, based on

the LS of the main predicator.
be-during0 (insurrection, [do0 (CNN, [interview0 (CNN, Max)])])
b. Step 2: Determine the actor and undergoer assignments, following the
Actor-Undergoer Hierarchy (AUH).
c. Step 3: Determine the morphosyntactic coding of the arguments

(privileged syntactic argument (PSA) selection, case assignment,
adposition assignment).
d. Step 4: Select the syntactic template(s) for the sentence following the
template selection principles.
e. Step 5: Assign arguments to positions in the syntactic representation of the

sentence.

Figure 19.15 Bock and Levelt’s model of grammatical encoding and RRG
There are five steps in the linking algorithm. It was pointed out in Van
Valin (2006: 280–282) that these five steps parallel the process of grammat-
ical encoding in the model of a speaker proposed in Bock and Levelt (1994).
This parallelism between a psycholinguistic model of speech production and
an abstract grammatical theory, as in Figure 19.15, suggests strongly that
the RRG semantics-to-syntax linking system has great potential as the basis
for a production model.
The steps in (18) involve applying rules (‘assign macroroles’, ‘select the
PSA’, ‘assign case’, etc.), selecting a bare syntactic template, and inserting
lexical items into the template. The content of these rules, etc., is the same
information contained in the linking templates and morphosyntactically
augmented LSs in the comprehension system, and this is no accident, of
course. As an alternative to applying these linking steps sequentially, it is
possible to condense the linking into constructing the semantic representa-
tion using the augmented LSs with lexical choices for the argument

Figure 19.16 Condensed semantics-to-syntax linking
positions in them and operators, selecting the passive linking template in

Figure 19.7, and integrating the two. This can be represented as in
Figure 19.16.
The semantic representation is composed of two LSs, do0 (A: x, [interview0
(x, U: y)]) and be-during0 (w, z), the lexical items filling the argument
positions in them (the LS for interview fills the z-argument position in the
prepositional LS), and the values for the minimally required operators. It is
integrated into the passive template from Figure 19.7 using the annotations
on the linking template: the undergoer Max occupies the core-initial PSA
position which carries the undergoer annotation, the actor CNN appears as
the object of the preposition by in the core-level periphery due to its actor
annotation, and the adjunct temporal preposition during heads a prepos-
itional phrase which is also in the core-level periphery. The general features
of the construction, that is, those that are cross-linguistically valid, as well as
the language-specific idiosyncratic features, the verb morphology, for
example, are contained in the constructional schema associated with the
passive templates. Thus, the five steps in the linking algorithm in (18) have
been reduced substantially into a fast production process, analogous to the
reduced comprehension process described in Section 19.4.
In (18) the AUH, the PSA Selection Hierarchy and the adposition assign-
ment rules, among others, play a role in the grammar version of the
linking algorithm. They represent important, linguistically significant
generalizations captured in RRG. Are they lost in the condensed linking
in Figure 19.16? They are not lost, but their status has changed. The AUH,
for example, governs macrorole assignment in step 2 in (18). There is no
macrorole assignment per se in the process in Figure 19.16: both the
linking template and the LS of the main predicator have the macroroles

specified in them. The linking regularities captured by the AUH are gener-
alizations across LSs; in other words, the generalizations expressed in the
AUH are now meta-generalizations across LSs (as in the meta-grammar of (L)
TAG; see Kallmeyer 2010) rather than being a rule constraining the assign-
ment of macroroles. The LSs in (3) express the basic relationship with an M-
transitive verb, while the LSs in (4), (5) and (11) lead to the meta-
generalization that variable undergoer choice is possible with certain
three-place verbs in certain languages. This also holds true for PSA selec-
tion, adposition and case assignment, etc. Thus, the constraints and rules
as well as the hierarchies that inform them are not invoked as grammat-
ical operations to be applied sequentially, as in (18), but rather are concep-
tualized as abstract meta-generalizations across the linking templates and
the augmented LSs.
One of the daunting challenges facing language processing models, both
of comprehension and production, is accounting for the incremental nature
of processing. This issue was addressed in Section 19.4 for comprehension,
where the solution was to posit parallel template processing, whereby the
number of competing templates is reduced as the system encounters more
and more of the input utterance, until eventually there is a single templatic
structure remaining (for unambiguous sentences). With production, on the
other hand, the issue is that speakers can start an utterance without having
planned it to the end (Bock 1995), and accordingly, the linking in (18) is an
idealization which assumes that the entire utterance is planned out in
advance. In reality, the semantic representation in Figure 19.16 is compat-
ible with the possibilities in (19), among others.
(19) a. CNN interviewed Max during the insurrection

b. During the insurrection, CNN interviewed Max.
c. During the insurrection CNN interviewed Max.
d. Max was interviewed by CNN during the insurrection.
e. During the insurrection, Max was interviewed by CNN.
f. During the insurrection Max was interviewed by CNN.
The speaker has to decide whether the main clause is to be active or passive
voice, a decision influenced by the relative topicality of the referents of the
RPs CNN and Max, and whether the temporal PP is to occur in its default
position in the core-level periphery or in initial position, and if initial,
whether it is a frame-setting topic in the pre-detached position or part of
the assertion in the pre-core slot. If the speaker begins with during, then
(19a, d) are eliminated, but the remaining four are still possible. If they
begin with CNN, then only (19a) is possible, and likewise if the utterance
starts with Max, only (19d) is possible. There are different commitment
points in the sentence; if it starts with the PP, the speaker is not yet
committed to the voice of the main clause, whereas if it starts with an
RP, the speaker is committed to the form of the main clause and the
placement of the PP.

This example has assumed that the speaker had decided on the predicator,
but it is not unusual for there to be competing predicators with different
implications in play. An alternative way of describing Max’s encounter with
CNN would be with the verb talk (to) as in (20). (The tense has been changed
to present perfect in both semantic representations.)
(20) a. Max has talked to CNN during the insurrection.

b. ⟨IF DEC ⟨TNS PRES ⟨ASP PERF [be-during0 (insurrection, [do0 (A: Max, [talk0
(Max, NMR to: CNN)])])]⟩⟩⟩
The speaker has options for the semantic representation as well as the
syntactic form of the utterance, and for the sake of the discussion, it is
assumed that Max is the topic and so the topic expression, the RP Max,
will be the subject of the main clause. Consequently, the main clause
beginning with Max has. . . can be continued with either been interviewed by
CNN or talked to CNN, reflecting two very different syntactic patterns. The
speaker can begin a sentence before deciding on a final structure for
the utterance.
Because structural templates play a crucial role in the processing system,
the incremental nature of processing is captured by competition among
possible templates which is resolved at critical points during the sentence.
For comprehension, the number of potentially relevant templates is reduced
word by word as the receiver processes the sentence linearly. The result of
this winnowing should ideally be a single structure associated with a single
meaning, and multiple structures and meanings create ambiguity. For
production, the speaker can begin uttering a sentence without having
decided the final form and meaning of it. Here again, there are semantic
representations and structural patterns competing, and the choices made at
critical points are a function of the speaker’s intention to communicate a
particular message.
It has become clear that in order to use an abstract competence grammar
for language processing, both production and comprehension, it is neces-
sary to make changes to the system. In particular, the rules and constraints
that the abstract grammar employs in the mapping between meaning and
form and between form and meaning are precompiled in the syntactic
templates and LSs for verbs and other predicators. These linking templates
and augmented LSs play a central role in both production and comprehen-
sion, affording rapid and direct coding in the case of production and what
native speakers perceive as nearly instantaneous interpretation with respect
to comprehension. There are, accordingly, two versions of RRG: the familiar
abstract grammar in Figure 19.17, and the processing system in Figure 19.18.
The same rules, principles, constraints, etc. are captured in both versions,
albeit differently. In the version in Figure 19.17, they are stated explicitly as
part of the linking algorithm, as in (18), whereas in the version in
Figure 19.18 they are meta-generalizations over the linking templates and
augmented LSs.

Figure 19.17 RRG as an abstract model of grammar
Figure 19.18 RRG as a language processing system
The double-headed arrow in Figure 19.18 signals that the linking is bidirec-
tional, which is expected in RRG, and it is annotated with ‘Winnowing’ and
‘Discourse-Pragmatics’. ‘Winnowing’ was discussed above. ‘Discourse-prag-
matics’ indicates that, from an RRG perspective, it is an important factor in
the mapping between form and meaning (see Bentley, Chapter 11, Latrouite
and Van Valin, Chapter 12). The interaction with morphosyntax has been
labelled ‘information structure’ and ‘information packaging’, and the latter
term refers to the packaging of the informational content of the utterance by
the speaker in order to facilitate its apprehension by interlocutors. This
entails that discourse-pragmatics affects the linking to syntax more than
the linking to semantics, because the speaker is making choices regarding
everything from lexical choices to word order which can be influenced by it.
Indeed, in Van Valin and LaPolla (1997: 423–430) it was shown that discourse-
pragmatics can interact with and influence every step of the semantics-to-
syntax linking (see also Sections 12.5–12.7 in Chapter 12). It is not the case
that it influences every step in every language; rather, looking across lan-
guages one finds the influence at a given step in one or more languages.
However, there are no steps in Figure 19.18. In terms of the processing version
of RRG, the influence of discourse-pragmatics can be seen in the lexicon and
the LTI. In the lexicon, it affects the choice of LS for the predicator, the choice
of referring expression(s) to fill the argument slots in the LS(s), and even in
some cases macrorole selection. In the LTI, it can influence the choice of

construction, whether the utterance needs a PrDS for a frame-setting topic, a

PrCS for a contrastive topic or focus, or the order of the words in the sentence,
among other things.
The importance of discourse-pragmatics for interpretation is rather differ-
ent. It is not tied to specific steps in the linking, even in the system in
Figure 19.17 (Van Valin and LaPolla 1997: 428). It ranges from constraining
the interpretation of pronouns within sentences (Van Valin and LaPolla
1997: §5.6), as well as the scope of quantifiers (Van Valin 2005: §3.6) and
negation (Van Valin and LaPolla 1997: 219), to supplying missing arguments
in discourse (Van Valin 2005: 171–175) and even missing predicators (Shi-
mojo 2008). It can, therefore, filter out certain possible interpretations, as
well as supporting other ones, as part of the winnowing.
19.7 Conclusion
This investigation of language processing in the brain from an RRG perspec-

tive took as its starting point a puzzling phenomenon: the ability of some
split-brain patients to make grammaticality judgements using only their
right hemisphere. The explanation for this phenomenon led to a view of
grammatical knowledge as being split up among linking templates, lexical
entries for verbs and other predicates, and the neural mechanisms in the
left temporal lobe which combine and match templates with sentences and
integrate the semantic representations in order to capture the basic ‘who
did what to whom’ meaning rapidly and efficiently (‘good enough’ inter-
pretation). Moreover, these same constructs showed themselves to be crucial
for language production as well. The title of Baynes and Gazzaniga (1988)
asks ‘Right hemisphere language: Insight into normal language mechan-
isms?’, and the answer is clearly affirmative.
On the RRG view, then, there is actually no ‘phrase structure building
process’ of the kind discussed in Section 19.2; the basic phrase structure
configurations already exist in the form of the linking templates, which
carry syntactic and semantic information, are combined to form more
complex structures, and integrate with the lexical semantic representations
of verbs and other predicates, which also carry morphosyntactic informa-
tion. Hence syntactic processing and semantic interpretation are intimately
intertwined, leading to the conclusion of Fedorenko et al. (2020: 1) that
‘lexico-semantic and syntactic processing are deeply interconnected and
perhaps not separable’.
The RRG conception of grammatical knowledge as involving linking tem-
plates turned out to have unexpected implications for the understanding of
certain types of aphasia. As discussed, there is a similar puzzling phenom-
enon in that domain, namely, the dissociation between the ability to make
grammaticality judgements and the inability to make use of that same
grammatical knowledge in comprehension and production. The

explanation for the dissociation turned out to involve the same devices as
the right hemisphere grammaticality judgements. While there were only a
small number of split-brain patients and only a small minority exhibited
any linguistic capacity in their right hemisphere, the dissociation in aphasia
has been documented in numerous studies involving speakers from a var-
iety of languages.
The paper began with a reference to Kaplan and Bresnan’s assertion that
there should be a link between abstract competence grammars and process-
ing models. It has been argued that the mechanisms of RRG can be the basis
for neurolinguistic models of both language comprehension and produc-
tion, and the exact nature of the adaptations required by the processing
models have been articulated explicitly and summarized in Figures 19.17
and 19.18. A distinctive characteristic of the RRG account is the distributed
representation of grammatical knowledge in terms of linking templates and
augmented LSs, which, in terms of the dual-stream models, is fully compat-
ible with them being accessible for the dorsal stream (production) as well as
the ventral stream (comprehension). RRG thus meets the requirement that a
grammatical theory claiming to capture speakers’ linguistic competence
should be able to be used in a testable psycholinguistic, or in this case,
neurolinguistic, processing model.
References
Baynes, Kathleen and Michael S. Gazzaniga. 1988. Right hemisphere lan-

guage: Insight into normal language mechanisms? In F. Plum (ed.), Lan-
guage, Communication and the Brain, 117–126. New York: Raven Press.
Blank, Idan, Zuzanna Balewski, Kyle Mahowald and Evelina Fedorenko.
2016. Syntactic processing is distributed across the language system.
NeuroImage 127: 307–323.
Bock, Kay. 1995. Sentence production: From mind to mouth. In J. L. Miller
and P. D. Eimas (eds.), Speech, Language and Communication (2nd ed.),
181–216. New York: Academic Press.
Bock, Kathryn and Willem Levelt. 1994. Language production: Grammatical
encoding. In M. Gernsbacher (ed.), Handbook of Psycholinguistics, 945–984.
Boland, Julie, Michael Tanenhaus, Susan Garnsey and Greg Carlson. 1995.
Verb argument structure in parsing and interpretation: Evidence from wh-
questions. Journal of Memory and Language 34: 774–806.
Bornkessel, Ina and Matthias Schlesewsky. 2006. The extended Argument
Dependency Model: a neurocognitive approach to sentence comprehen-
sion across languages. Psychological Review 113: 787–821.
Bornkessel-Schlesewsky, Ina and Matthias Schlesewsky. 2013. Reconciling
time, space and function: A new dorsal–ventral stream model of sentence
comprehension. Brain and Language 125: 60–76.

Brennan, Jonathan, Yuval Nir, Uri Hasson, Rafael Malach, David J. Heeger
and Liina Pylkkänen. 2012. Syntactic structure building in the anterior
temporal lobe during natural story listening. Brain and Language 120:
163–173.
Brodmann, Korbinian. 1909. Vergleichende Lokalisationslehre der Grosshirnrinde.
Leipzig: Johann Ambrosius Barth.
Clifton, Charles, Jr. and Lynn Frazier. 1989. Comprehending sentences with
long-distance dependencies. In Greg Carlson and Michael Tanenhaus
(eds.), Linguistic Structure in Language Processing, 273–317. Dordrecht:
Kluwer.
Davis, Anthony and Jean-Pierre Koenig. 2000. Linking as constraints on word
classes in a hierarchical lexicon. Language 76: 56–91.
Dronkers, Nina F., David P. Wilkins, Robert D. Van Valin, Jr., Brenda B.
Redfern and Jeri J. Jaeger. 1994. A reconsideration of the brain areas
involved in the disruption of morphosyntactic comprehension. Brain and
Language 47(3): 461–463.
Dronkers, Nina F., David P. Wilkins, Robert D. Van Valin, Jr., Brenda B.
Redfern and Jeri J. Jaeger. 2004. Lesion analysis of the brain areas involved
in language comprehension. Cognition 92: 145–177.
Federenko, Evelina, Idan A. Blank, Mathew Siegelmann and Zachary Miner-
off. 2020. Lack of selectivity for syntax relative to word meanings through-
out the language network. Cognition 203: 1–19.
Ferreira, Fernanda and Nikole D. Patson. 2007. The ‘good enough’ approach
to language comprehension. Language and Linguistics Compass 1(1–2): 71–83.
Friederici, Angela D. 2009. Pathways to language: Fiber tracts in the human
brain. Trends in Cognitive Sciences 13: 175–181.
Friederici, Angela D. 2012. The cortical language circuit: From auditory
perception to sentence comprehension. Trends in Cognitive Sciences 16:
262–268.
Friederici, Angela D., Martin Meyer and D. Yves von Cramon. 2000. Auditory
language comprehension: An event-related fMRI study on the processing
of syntactic and lexical information. Brain and Language 74: 289–300.
Friederici, Angela D., Shirley-Ann Rüschemeyer, Anja Hahne and Christian J.
Fiebach. 2003. The role of left inferior frontal and superior temporal
cortex in sentence comprehension: Localizing syntactic and semantic
processes. Cerebral Cortex 13: 170–177.
Gazzaniga, Michael S. 1983. Right hemisphere language after brain bisec-
tion: A 20-year perspective. American Psychologist 38: 525–537.
Gazzaniga, Michael S. 2000. Cerebral specialization and interhemispheric
communication: Does the corpus callosum enable the human condition?
Brain 123: 1293–1326.
Gazzaniga, Michael S. 2005. Forty-five years of split-brain research and still
going strong. Nature Reviews Neuroscience 6: 653–659.
Gazzaniga, Michael S. and Steven A. Hillyard. 1971. Language and speech
capacity of the right hemisphere. Neuropsychologia 9: 273–280.

Herrmann, Björn, Burkhard Maess, Anja Hahne, Erich Schröger and

Angela D. Friederici. 2011. Syntactic and auditory spatial processing
in the human temporal cortex: An MEG study. NeuroImage 57: 624–633.
Hickok, Gregory and David Poeppel. 2004. Dorsal and ventral streams:
A framework for understanding aspects of the functional anatomy of
language. Cognition 92: 67–99.
Hickok, Gregory and David Poeppel. 2007. The cortical organization of
speech perception. Nature Reviews Neuroscience 8: 393–402.
Jurafsky, Dan. 1996. A probabilistic model of lexical and syntactic access and
disambiguation. Cognitive Science 20: 137–194.
Kallmeyer, Laura. 2010. Parsing Beyond Context-Free Grammars. Heidelberg:
Springer.
Kallmeyer, Laura, Rainer Osswald and Robert D. Van Valin, Jr. 2013. Tree
wrapping for Role and Reference Grammar. In G. Morrill and M.-J. Neder-
hof (eds.), Formal Grammar: Proceedings of the 17th and 18th International
Conferences (Lecture Notes in Computer Science 8036), 175–190. Heidelberg:
Springer.
Kaplan, Ronald and Joan Bresnan. 1982. Lexical-Functional Grammar:
A formal system for grammatical representation. In Joan Bresnan (ed.),
The Mental Representation of Grammatical Relations, 173–281. Cambridge, MA:
MIT Press.
Koenig, Jean-Pierre and Gail Mauner. 2003. Arguments for adjuncts. Cognition
89: 67–103.
Lau, Ellen F., Colin Phillips and David Poeppel. 2008. A cortical network
for semantics: (de)constructing the N400. Nature Review Neuroscience 9:
920–933.
Linebarger, Marcia C., Myrna F. Schwartz and Eleanor M. Saffran. 1983.
Sensitivity to grammatical structures in so-called agrammatic aphasics.
Cognition 13: 361–392.
Lu, Ching-Ching, Elizabeth Bates, Ping Li, Ovid Tzeng, Daisy Hung, Chih-Hao
Tsai et al. 2000. Judgements of grammaticality in aphasia: The special case
of Chinese. Aphasiology 14: 1021–1054.
Lukatela, Katerina, Stephen Crain and Donald Shankweiler. 1988. Sensitivity
to inflectional morphology in agrammatism: Investigation of a highly
inflected language. Brain and Language 33: 1–15.
Magnusdottir, S., P. Fillmore, D. B. den Ouden, H. Hjaltason, C. Rorden, O.
Kjartansson, et al. 2013. Damage to left anterior temporal cortex predicts
impairment of complex syntactic processing: A lesion-symptom mapping
study. Human Brain Mapping 34: 2715–2723.
Mazoyer, B. M., N. Tzourio, V. Frak, A. Syrota, N. Murayama, O. Levrier, G.
Salamon, S. Dehaene, L. Cohen and J. Mehler. 1993. The cortical represen-
tation of speech. Journal of Cognitive Neuroscience 5: 467–479.
McKoon, Gail and Jessica Love. 2011. Verbs in the lexicon: Why is hitting
easier than breaking? Language and Cognition 3(2): 313–330.

Nolan, Brian and Elke Diedrichsen (eds.). 2013. Linking Constructions into
Functional Linguistics: The Role of Constructions in Grammar. Amsterdam: John
Benjamins.
Obleser, Jonas, Jonas Zimmermann, John Van Meter and Josef P.
Rauschecker. 2007. Multiple stages of auditory speech perception reflected
in event-related fMRI. Cerebral Cortex 17: 2251–2257.
Pylkkänen, Liina. 2019. The neural basis of combinatory syntax and seman-
tics. Science 366: 62–66.
Saur, Dorothee, Björn W. Kreher, Susanne Schnell, Dorothee Kümmerer,
Philipp Kellmeyer, Magnus-Sebastian Vry, et al. 2008. Ventral and dorsal
pathways for language. PNAS USA 105(46): 18035–18040.
Schlesewsky, Matthias and Ina Bornkessel. 2004. On incremental interpret-
ation: Degrees of meaning accessed during sentence comprehension. Lin-
gua 114: 1213–1234.
Schwartz, Myrna F., Eleanor M. Saffran and Oscar S. Marin. 1980. The word
order problem in agrammatism: I. Comprehension. Brain and Language 10:
249–262.
Schwartz, Myrna F., Marcia C. Linebarger, Eleanor M. Saffran and David S.
Pate. 1987. Syntactic transparency and sentence interpretation in aphasia.
Language and Cognitive Processes 2: 85–113.
numeral quantifier construction in Japanese. In Robert Van Valin (ed.),
Investigations of the Syntax–Semantics–Pragmatics Interface, 285–304. Amster-
Shtyrov, Yury, Friedemann Pulvermüller, Risto Näätänen and Risto J. Ilmo-
niemi. 2003. Grammar processing outside the focus of attention: An MEG
study. Journal of Cognitive Neuroscience 15: 1195–1206.
Stowe, L. A. 1985. Parsing wh-constructions: Evidence for on-line gap loca-
tion. Language and Cognitive Processes 1: 227–245.
Stowe, L. A., C. A. J. Broere, A. M. J. Paans, A. A. Wijers, G. Mulder, W.
Vaalburg and F. Zwarts. 1998. Localizing cognitive components of a
complex task: sentence processing and working memory. Neuroreport 9:
2995–2999.
Stromswold, Karin, David Caplan, Nathaniel Alpert and Scott Rauch. 1996.
Localization of syntactic comprehension by positron emission tomog-
raphy. Brain and Language 52: 452–473.
Townsend, David J. and Thomas G. Bever. 2001. Sentence Comprehension: The
Integration of Habits and Rules. Cambridge, MA: MIT Press.
Traxler, M. and M. Pickering. 1996. Plausibility and the processing of
unbounded dependencies: An eye-tracking study. Journal of Memory and
Language 35: 454–475.
Turken, And U. and Nina F. Dronkers. 2011. The neural architecture of the
language comprehension network: Converging evidence from lesion and
connectivity analyses. Frontiers in Systems Neuroscience 5: 1.

Tyler, L. K. and W. Marslen-Wilson. 2008. Fronto-temporal brain systems

supporting spoken language comprehension. Philosophical Transactions of
the Royal Society B: Biological Sciences 363: 1037–1054.
Van Valin, Robert D., Jr. 2006. Semantic macroroles and language processing.
In Ina Bornkessel, Matthias Schlesewsky, Bernard Comrie and Angela D.
Friederici (eds.), Semantic Role Universals and Argument Linking: Theoretical,
Typological, and Psycholinguistic Perspectives, 263–301. Berlin: Mouton de
Gruyter.
Wang, Luming, Matthias Schlesewsky, Balthasar Bickel and Ina Bornkessel-
Schlesewsky. 2009. Exploring the nature of the ‘subject’-preference: Evi-
dence from the online comprehension of simple sentences in Mandarin
Chinese. Language and Cognitive Processes 24: 1180–1226.
Wulfeck, Beverly B. 1988. Grammaticality judgments and sentence compre-
hension in agrammatic aphasia. Journal of Speech and Hearing Research 31:
72–81.
Wulfeck, Beverly, Elizabeth Bates and Rita Capasso. 1991. A crosslinguistic
study of grammaticality judgments in Broca’s aphasia. Brain and Language
41: 311–336.
Zaidel, Eran. 1983. On multiple representations of the lexicon in the brain:
The case of two hemispheres. In Michael Studdert-Kennedy (ed.), The
Psychobiology of Language, 105–122. Cambridge, MA: MIT Press.
Notes
1 See Gazzaniga (2005) for an overview of the history of split-brain research

and (2000) for an explication of the methodologies used.
2 Brodmann (1909) analysed the structure of the cortex and grouped areas
in the cortex with the same cytoarchitectonic features into what are
called ‘Brodmann’s areas’. This means that all of the cortex in a given
BA, e.g. BA 22, has the same cytoarchitecture.
3 Some models, e.g. Hickok and Poeppel (2007), Tyler and Marslen-Wilson
(2008), posit right-hemisphere involvement in the comprehension process,
and these concern semantic and pragmatic aspects of comprehension, not
morphosyntactic processing, which they locate exclusively in the
left hemisphere.
4 Pylkkänen (2019) proposes a view of language in the brain which crucially
involves the left anterior temporal lobe and which treats ‘syntax as
knowledge, semantics as process’ (64–65):

The neuroscience-of-language field has long assumed that our brains

build syntactic structure during language processing. Today it is rea-
sonable to question this assumption. . . . But what if we cannot find
evidence that our brains actually build syntactic structure? Syntax
may be something that the brain knows rather than does. Perhaps
the combinatory steps, which consume energy and make our neurons
fire, are all semantic, and syntactic processing amounts to comparing
these semantic structures to our stored knowledge of syntax. The
knowledge may have the format of generative rules that create struc-
ture . . . or may represent the structures themselves.
5 This should not be interpreted to imply that no other brain regions in

the left temporal lobe are involved in the comprehension process; the
point is that BA 22 is important for it.
6 Be is attached to the AUX node in the constituent projection because it is
structurally required for non-verbal nuclei and passive verbs as well,
whereas ‘(aux)’ in the templates represents auxiliary verbs which are
only part of the expression of operators, e.g. be in the progressive aspect
or have in the perfect. Structural be is required even when there are no
operators, e.g. Sally wants to be a lawyer/tall.
7 The precompiling of syntactic and semantic information as proposed
in Van Valin (2006) and here moves RRG closer to (Sign-based) Con-
struction Grammar and Head-Driven Phrase Structure Grammar (e.g.
Davis and Koenig 2000). RRG was already a species of construction
grammar in the generic sense, since it recognizes the importance and
centrality of the notion of grammatical construction in linguistic
theory and description; the realization of this in the theory is rather
different from ‘standard’ versions of Construction Grammar,
however.
8 This should be read as: in the linking template the RP the burglar is the A,
and in the LS the A is the x-argument, and consequently the burglar fills
the x-argument position in the LS. The same holds for the window and the
y-argument position in the LS.
9 Did is in curly brackets to signify that it is part of the operator projection
rather than the constituent projection.
10 The embedded clause can be iterated for additional long-distance wh-
questions. The long-distance (non-local) nature of the tree in Figure 19.11
is an unusual feature for templates, since in approaches like (L)TAG
(Kallmeyer 2010) tree templates are all local, and a non-local tree would
be the result of multiple combinatory operations (Kallmeyer et al. 2013).
For an explanation of why the embedded clause is a daughter of the
CLAUSE node rather than the CORE node in Figure 19.11, see Van Valin
2005: §6.3. The tree in Figure 19.12 also has some non-local properties,
albeit different ones from Figure 19.11.

11 This account is somewhat oversimplified, but it is adequate for the

purposes of this discussion.
12 There are languages with no adjacency restrictions on RP constituents
and no grammatical constraints on linearization, e.g. many Australian
Aboriginal languages, and they too would have unordered clausal and
phrasal templates, yielding grammatically (but not pragmatically)
unconstrained word order. See Van Valin (2005: 28–29) for an example.
13 The presence of word storage in the right hemisphere raises interesting
questions for theories like Distributed Morphology, which do not posit a
lexicon in the usual sense but rather a repository of abstract roots that
must undergo derivation in the syntax before they can surface as words.
If this repository were in both hemispheres, the isolated right hemi-
sphere in a split-brain patient would have no access to the syntactic
machinery which derives words from the abstract roots. Yet some are
able to recognize words, which is unexpected due to the lack of the
necessary syntactic machinery.
14 In an RRG grammar these constraints would be expressed in the
constructional schema for the construction. We refer to the relevant discus-
sion in Chapter 1 (§§1.6.3–1.6.4) and to Van Valin (2005: Chapters 5, 7) and
Nolan and Diedrichsen (2013) for extensive exemplification.
15 All of the values except the ones for mean accuracy are extrapolated
from Fig. 1 in Baynes and Gazzaniga (1988).
16 This actually involves both lexical and phrase-structural information,
both of which are available in the right hemisphere, as noted above,
since the lexical entries for verbs are the same in both hemispheres.
17 These included: gaps associated with wh-questions (subject and object),
missing arguments in infinitival but not tensed complements,
double gaps.
18 How many birds did you see? vs. *How many did you see birds?
19 It has been suggested that J.W.’s right-hemisphere grammatical ability is
the result of an incomplete isolation of the hemispheres or of incom-
plete left-lateralization of language at the time of the surgery. The
former is unlikely, for three reasons. First, Baynes and Gazzaniga report
that an MRI showed a complete separation of the hemispheres, second,
they also point out that there is no evidence of cross-cueing between the
hemispheres, and third, his test results were different from V.P.’s, who
did not have a complete isolation of the two hemispheres. With respect
to the possibility of incomplete lateralization, it can’t be ruled out a
priori, but it predicts neither the pattern of his responses nor the simi-
larity of his responses to the aphasics, unlike the account
proposed above.

20
Formalization of
RRG Syntax
Laura Kallmeyer and Rainer Osswald
Abbreviations

AC actor MOD modal

ACT active NUC nucleus
AM accompanied motion OP operator
ANIM animate OPS operator structure
AOR aorist PERIS periphery structure
ASP aspect PrCS pre-core slot
CFTG Context-Free Tree Grammar PSBL possible
CO core RP reference phrase
DP detached position STA status
ECS extra-core slot TAG Tree Adjoining Grammar
EDL extended domain of locality TNS tense
IF illocutionary force TWG Tree Wrapping Grammar
LCFRS Linear Context-Free Rewriting UG undergoer
System
LM linkage marker
20.1 Introduction
Role and Reference Grammar (RRG) has been developed as a theory of

grammar that covers typologically distinct languages and is able to capture
the interaction between syntax, semantics and pragmatics.* The design of
* The research presented in this chapter was supported by the Collaborative Research Centre 991 ‘The Structure of
Representations in Language, Cognition, and Science’ funded by the German Research Foundation (DFG) and the project
TreeGraSP funded by the European Research Council (ERC).

738 LAURA KALLMEYER AND RAINER OSSWALD
RRG was not driven by specific formal considerations. In particular, there is

no formal core that plays a crucial role in RRG, as, for example, the theory of
feature structures does in Head-Driven Phrase Structure Grammar (HPSG;
Pollard and Sag 1994). The goal of this chapter is to present RRG as a
formalized grammatical theory that puts emphasis on mathematical and
logical rigour. In particular, it will be shown how the possible universal
and language-specific syntactic templates of RRG can be formally specified,
and by which formal operations they can be combined into larger
syntactic structures.
Working typologists who use RRG as their framework for linguistic analysis
may not regard a thorough formalization as particularly important. In fact,
one of the appeals RRG has for field linguistics is that it does not come with an
overly heavy theoretical load but keeps a good balance between a rich and
elaborate set of notions and explanatory mechanisms and a semi-formal,
intuitive presentation. A formalization can, however, help to identify and
eliminate possible gaps and inconsistencies of the theory and, thereby, to
improve the theory. Moreover, a formalization can serve as a basis for compu-
tational implementations of RRG. While a thorough formalization may not be
absolutely necessary for a computational treatment from an engineering
perspective (see Chapter 21 on computational implementation and applica-
tions of RRG), it can contribute to implementations that give full consider-
ation to the overall architecture of RRG as a theory of grammar.
20.2 The Task of Formalizing RRG Syntax
In RRG, syntactic representations are composed of syntactic templates

stored in the syntactic inventory. Figure 20.1a shows a simple example of a
syntactic representation; possible candidates for syntactic templates are
shown in Figure 20.1b. A formalization first needs to decide on what kind
of formal structures to use. In line with the tradition of RRG, tree structures
will be employed for this purpose, where nodes can carry additional features
(a) (b)
SENTENCE SENTENCE
CLAUSE CLAUSE
PrCS CORE PERIPHERY CLAUSE
NUC PrCS CORE CORE PERIPHERY
RP RP PRED CORE ADV

PrCS
V ADV RP NUC
RP PRED
what did Kim smash yesterday
V
TNS CLAUSE
Figure 20.1 Examples of syntactic representation and syntactic templates in RRG

Formalization of RRG Syntax 739
besides the category labels. The second task is to define the modes by which
the syntactic representations are composed from the members of the syn-
tactic inventory (Section 20.3). Next, the templates available in the inventory
must be specified, and this should be done in a way that allows us to capture
generalizations among them within and across languages (Section 20.4).
A further point that needs special treatment is the formalization of the
operator projection (Section 20.5).
A key component of RRG’s approach to syntactic analysis is the layered
structure of the clause: sentences are assumed to have an internal structural
layering consisting of clause, core and nucleus. The different layers serve as
attachment sites for different types of operators: tense operators attach to
the clause, modality to the core, aspect to the nucleus, etc. The core level is
also the default attachment site for arguments. In the following, we will
refer to the subtree of a syntactic representation consisting of the root and
its non-peripheral clause, core, nucleus and predicating descendants as the
clausal skeleton of the representation. The syntactic structures in RRG are
basically labelled trees, and there are good reasons to use tree structures in a
formalization as well. Trees provide the most natural way to analyse syntac-
tic structures since they build on the basic relations of immediate dominance
and linear precedence.
20.2.1 Approaches to Formalizing RRG Syntax

The formal specification of syntactic structures in RRG is briefly addressed
by Van Valin and LaPolla (1997, §2.5), where two approaches are discussed:
(i) the specification by immediate dominance and linear precedence rules
and (ii) the specification by syntactic tree templates and their combination.
As to approach (i), Van Valin and LaPolla (1997: 69f ) propose the following
universal immediate dominance rules for the constituent structure of
simple sentences:1
(1) SENTENCE ! {(DP)}, CLAUSE

DP ! XP j ADV
CLAUSE ! {(ECS)}, CORE, (PERIPHERY), {NP*}
ECS ! XP j ADV
PERIPHERY ! XP j ADV
CORE ! ARG*, NUC
NUC ! PRED
PRED ! V j XP
ARG ! PRO j XP
XP ! NP j PP
The commas on the right-hand side of these rules do not indicate any
ordering of the subconstituents. The ordering is specified by additional
linear precedence rules, which are partly universal and partly language-
specific. For example, English, a verb-medial language, obeys the following
linear precedence rules (Van Valin and LaPolla 1997: 71):

(2) NP > NUC

NUC > NP* > PP*
As Van Valin and LaPolla point out, the range of possible syntactic constella-
tions specified by the above rules needs to be further constrained by the
linking of syntax to semantics, which includes constraints on the syntactic
realization of arguments depending on semantic structures in the lexicon.
In addition to the constituent structure rules in (1) and (2), a separate set of
immediate dominance and linear precedence rules is needed for specifying
the structure of the operator projection.
The idea of specifying the constituent structure and the operator structure
separately by different context-free grammars was originally proposed by
Johnson (1987), based on the observation that the ordering among the
operators is systematically correlated with their scope given by their attach-
ment site at the clausal skeleton, whereas the surface order of the operators
relative to arguments and adjuncts is much less transparent and often
requires crossing branches. The two grammars taken together then consti-
tute a projection grammar, giving rise to a constituent projection and an operator
projection. Johnson formally defines a projection grammar as a quadruple
P ¼ T; I; ðAi Þi2I , ðGi Þi2I , where T is a set of terminal symbols, I is a finite
set of ‘projection’ indices and, for every i 2 I, Ai is a subset of T and Gi is a
formal grammar with terminal symbols in Ai . A string s over T then belongs
to the language generated by P if and only if its ith projection, that is, the
concatenation of the elements of s belonging to Ai , is in the language
generated by Gi .
While it seems reasonable to distinguish between constituent structure
and operator structure, Johnson’s proposal has the problem of being purely
surface oriented. As a consequence, it does not enforce matching clausal
skeletons in the two projections. However, corresponding clausal skeletons
in both projections are taken for granted in the syntactic representations of
RRG. A further problem arises from the assumption that the operator
projection can be represented as a tree, that is, that each operator contrib-
utes only to one layer (cf. Section 20.5 for counterexamples).
Approach (ii), the second approach discussed by Van Valin and LaPolla,
postulates an inventory of elementary syntactic trees that can be combined
into more complex syntactic structures. Van Valin and LaPolla (1997: 654,
note 34) point out that Tree Adjoining Grammars (TAG, Joshi and Schabes
1997) may provide a way to formalize such tree templates and their
composition. Building on this idea, Kallmeyer et al. (2013) (see also Kall-
meyer and Osswald 2017; Osswald and Kallmeyer 2018) propose a formal-
ization of RRG as a grammar based on so-called elementary trees and TAG-
inspired tree composition operations. We will detail this approach in
Section 20.3.
A slightly different proposal has been suggested by Nolan (2004), who
argues for a formalization of RRG that systematically exploits feature-based

representations, similar in style to HPSG (Pollard and Sag 1994) and, more
recently, Sign-Based Construction Grammar (Sag 2012). Representing con-
stituent structures in this way calls for features, or attributes, by which the
subconstituents can be addressed. This can be done either by reconstructing
tree structures as feature structures based on formal features such as first
and rest, or by employing functional notions like subject, direct-object,
etc. However, in RRG, configurational syntactic notions are usually not
considered as basic but rather as derived concepts. While the representation
of the constituent projection proposed in Nolan (2004, §5.5), which builds on
the immediate dominance rules given in (1), is not fully explicit about the
attributes involved, it seems that it uses either pure list-oriented attributes
(first and rest, or even 1st, 2nd, 3rd, etc.) or attributes specific to the
categories of the subconstituents.
20.2.2 The Formal Treatment of Operators and

Peripheral Elements
The formalization presented in the following assumes a single syntactic tree
structure in which operator components are distinguished by a special
category OP and a feature structure that characterizes their contribution.
For example, the operator did in Figure 20.2 contributes [tense past] to the
CLAUSE and the definiteness operator the contributes [def þ] to the RP layer
of boy.2 This representation, together with the approach to operator adjunc-
tion presented in Section 20.5, turns out to be sufficient for capturing the
scope-related ordering among the operators. The operator projection can
then be defined as the subtree consisting of the clausal skeleton plus the
components with category OP.3
In RRG’s graphical presentations of syntactic structures, peripheral
structures and linkage markers are usually attached to the clausal
SENTENCE
CLAUSE
PrCS OP[TENSE past] CORE
RP did RP NUC
PRO OP[DEF +] CORER V[PRED +]
what the NUCR smash
boy
Figure 20.2 Operator marking by features

skeleton by arrows (cf. Figure 20.1a). The formalization proposed here

does not make use of separate PERIPHERY nodes but marks peripheral
structures by a feature [peri þ], as indicated in Figure 20.3. As with
operators, peripheral elements are subject to the iconicity principle that
their relative ordering respects the layering of their attachment sites. This
aspect will be taken care of in the formalization of the operator projection
and the periphery structure described in Sections 20.5 and 20.7. The
proposed modes of combining peripheral structures with the clausal skel-
eton are similar to those used for operators (Section 20.3.2). Linkage
markers can be treated similarly.
To sum up, the syntactic structures of RRG can be formalized as labelled
trees, where node labels can carry additional features. Since the labels can
be regarded as feature values, too (e.g. category labels as values of the feature
cat), we may assume without restriction of generality that node labels are
sets of attribute-value specifications. From this perspective, MP[PERI þ] is short
for [cat mp, peri þ]. Introducing features allowed us to get rid of the
PERIPHERY nodes, whose only purpose is to mark their daughters as periph-
eral. By the same line of reasoning, we can eliminate the PRED nodes, whose
purpose is to mark their daughters as predicating (cf. Van Valin 2005: 13), by
adding [PRED þ] to the respective daughter nodes, as illustrated by the
V nodes in Figures 20.2 and 20.3.
SENTENCE
CLAUSE
CORE
RP MP[PERI +] NUC RP MP[PERI +]
CORER COREM V[PRED +] OP[DEF +] CORER COREM
NUCR NUCM smashed the NUCR NUCM
N ADV MP[PERI +] N ADV
Kim deliberately COREM vase yesterday
NUCM
blue
Figure 20.3 Periphery marking by features

20.3 A Tree Rewriting Formalism for Syntactic

Composition in RRG
The standard presentation of RRG gives only an informal description of

how syntactic templates are combined to more complex syntactic struc-
tures. As mentioned in Section 20.2.1, Van Valin and LaPolla (1997: 654,
note 34) suggest that a formal account of the modes of composition may
show some similarity to Tree Adjoining Grammars (TAG, Joshi and Schabes
1997). They rightly point out that TAG is a grammar formalism and not a
linguistic theory in its own right. TAG per se does not make any commit-
ments about what kind of categories and what kind of syntactic configur-
ations are appropriate for linguistic analysis. There is one caveat, however:
the standard adjunction operation of TAG aims at modifying binary
branching structures. Since flat syntactic structures are prevalent in RRG,
the formalization proposed in the following employs slightly different
modes of composition.
20.3.1 Elementary Trees

The architecture of RRG assumes an inventory of syntactic templates as
elementary building blocks for syntactic representations (cf. Figure 20.1b).
These templates often have a more complex structure than just consisting of
a root node together with a series of immediate daughters. They can thus
capture a wider range of dependencies than standard phrase structure rules
like those listed in (1). In TAG, this property is called the extended domain of
locality (EDL) represented in elementary trees (Joshi and Schabes 1997: 95f ). In
particular, templates can contain a predicative node and slots for all the
arguments of that predicate. An example is given in Figure 20.4, which
shows the templates that could be used for generating the syntactic struc-
ture for (3).
(3) Kim smashed the blue vase.
In this example, the transitive verb template associated with the verbal
predicate smash contains not only the NUC node but also the CORE and
the two RP argument slots. Note that the lexical elements (marked grey in
the figure) are not part of the templates but are added in a separate step.
The availability of larger syntactic units allows one to lexicalize parts of the
grammar, up to the point of a fully Lexicalized Tree Adjoining Grammar
(LTAG), in which every elementary tree is required to have a lexical anchor.
The similarities between RRG’s tree templates and (L)TAG’s elementary
trees suggest a formalization of RRG syntax along the following lines:
A language is syntactically described by a tree rewriting grammar compris-
ing a set of tree templates, lexical elements filling the anchor nodes of these
templates, and certain operations for combining them into syntactic

CORE CLAUSE SENTENCE
RP RP NUC RP CORE CLAUSE
CORER PRED
NUCR V DEF RP
PERIPHERY
N smashed the MP CORER
Kim COREM NUCR
NUCM N
A vase
blue
Figure 20.4 RRG templates
representations of phrases and sentences. More specifically, a grammar

consists of a finite set of elementary trees that can be composed inductively
via three basic tree composition operations, namely sister adjunction, substitu-
tion and wrapping substitution. Elementary trees are defined as labelled
ordered trees whose internal nodes are labelled with categories such as
CLAUSE, CORE, RP, N, V etc., and whose leaf nodes are labelled either with
lexical items or with syntactic categories. The assumption that elementary
trees are ordered trees means: (i) every two nodes that do not stand in a
dominance relation are ordered by linear precedence; (ii) if a node n1
precedes a node n2 then every node dominated by n1 precedes every node
dominated by n2. Note that these requirements exclude structures with
crossing branches.
Figure 20.5 shows possible elementary trees (including lexical elements)
for the syntactic analysis of sentence (3) and illustrates how they combine.
The choice of the elementary trees in the example reflects the following
general principles: Each lexical predicate comes with its entire layered
structure, including argument slots for all its arguments (see the trees for
Kim, smashed, blue and vase in Figure 20.5). Periphery elements, operators and
linkage markers contribute an adjunct tree. Such a tree cannot fill an
argument slot but has to be adjoined, that is, attached, to the clausal
skeleton as an additional daughter of some appropriate node. An asterisk
at the root node indicates that a tree is an adjunct tree (see the trees for the
operator the and the modifier blue in Figure 20.5). In addition, there can also
be non-lexicalized elementary trees (see the SENTENCE–CLAUSE tree in
Figure 20.5).
The formalization of RRG syntax presented here allows for a further
level of (de)composition: Elementary trees are specified in a so-called

SENTENCE
SENTENCE
CLAUSE CLAUSE
CLAUSE
CORE
RP CORE
RP NUC RP
RP CORER RP NUC RP
V
RP∗
CORER CORE∗R NUCR CORER V OPdef CORER
smashed
OPdef
NUCR MPperi N NUCR smashed the MPperi NUCR
the
N COREM vase N COREM N
Kim NUCM Kim NUCM vase
A A
blue blue
Figure 20.5 Elementary RRG trees and their composition
CLAUSE RP CORE∗R
CORE CORER MPperi
V RP NUC RP NUCR COREM

N
smashed V N NUCM
vase
A
A
blue
Figure 20.6 Lexical anchoring of elementary tree templates
meta-grammar, a system of tree constraints, which captures generalizations

across elementary trees (more on this in Section 20.4.2). Moreover, lexical
items are stored in a separate lexicon and enter their elementary trees by a
process of lexical anchoring under constraints. In other words, we distin-
guish unanchored elementary trees from the anchored trees that enter tree
composition. The former are called elementary tree templates. Figure 20.6
shows the decomposition of three of the elementary trees from Figure 20.5
into their tree template and the lexical anchor. The place where the anchor
has to be inserted is marked with a diamond symbol.
Note that introducing elementary trees as ordered trees does not prevent
us from defining grammars for languages with free word order. If, for
instance, a language allows the arguments of a verb to appear in an arbitrary
order, then the verb would have elementary trees for each such ordering at
its disposal. What is important here is that elementary trees are not atomic

building blocks but are generated in a modular fashion from classes of tree
constraints in the meta-grammar. Since fewer ordering constraints in the
meta-grammar correspond to a larger set of elementary tree templates, it
follows, loosely speaking and without taking into account morphological
markings, that free word order is descriptively more simple in the meta-
grammar than strict word order.
20.3.2 Simple Substitution and Sister Adjunction

The most basic mode of composition for syntactic templates is substitution.
The trees for Kim and for vase in Figure 20.5, for instance, are added by
substitution to the tree of smashed, filling the two RP argument slots. In the
following, a tree with label X is meant to be a tree whose root carries the
category label X. A tree β with label X can be substituted for a leaf node
labelled with X (the substitution node) of a tree α by ‘identifying’ the root node
of β with the substitution node (cf. Figure 20.7a). More generally, if the nodes
are labelled by feature structures, the two feature structures must be com-
patible, and the node of the resulting tree is labelled by the unification of the
two feature structures. Each non-terminal leaf node in an elementary tree is
a substitution node and must obligatorily be filled by substitution or by the
substitution part of wrapping substitution (see 20.3.3). Substitution is the
main mode of composition for expanding argument nodes by the syntactic
representations of specific argument realizations.
In RRG, periphery elements and operators are only indirectly linked to the
layered structure, by a PERIPHERY edge or by an edge to their layer in the
operator projection. We include them in the layered constituent structure
while making sure the information about whether they are operator or
periphery elements is preserved. More concretely, the directed edge with
label PERIPHERY is replaced by an unlabelled immediate dominance edge
between the target layer and the periphery element and a feature [peri þ]
on the root of the peripheral structure, often abbreviated as a subscript peri
on the node category. In the case of operators, they also attach as a daughter
(a)
(b)
Figure 20.7 Schematic sketch of simple substitution (a) and sister adjunction (b)

to the target node in the constituent tree. Their category (OP) indicates that
they are operators and their contribution is indicated within further fea-
tures attached to the OP node. Their elementary trees are rooted by a node
that merges with the target tree. An advantage of this is that one can
formulate constraints by means of feature specifications. For instance, one
can require that there is only one definiteness operator in an RP. We will see
how to use features for this purpose in more detail in Section 20.5.
Peripheral structures cannot be added by substitution since they do not
attach to leaves but to internal nodes, in general, and the same holds for the
operators; cf. Figure 20.5. The mode of composition proposed for these cases
is (sister) adjunction (see also Kallmeyer et al. 2013).4 As with substitution, we
assume that the templates available for adjunction have a root label which
coincides with the label of the target node (cf. Figure 20.5). For convenience,
the root of an adjunction tree is marked by an asterisk in the graphical
presentations. We call elementary trees with this marking adjunct trees.
A further constraint on adjunct trees is that their root node has only a
single daughter. The idea behind this is, as explained earlier, that the tree
below the root is the actual periphery element while the root node captures
more or less the PERIPHERY edge from RRG textbooks. The root label of an
adjunction tree specifies the attachment site at the phrasal skeleton. In
Figure 20.8, for example, the adverbial completely adjoins at the nucleus
while yesterday adjoins at the core.
CLAUSE
CORE
RP RP NUC RP CORE∗
Kim NUC∗ V MPperi
MPperi RP
finished
yesterday
the cake
completely
CLAUSE
CORE
RP NUC RP MPperi
Kim MPperi Vpred the cake yesterday
completely finished
Figure 20.8 Sister adjunction of periphery elements at different layers

Sister adjunction is defined in such a way that the target node must
not be a leaf node. Because of their single-daughter property, roots of
adjunct trees are excluded as target nodes for sister adjunction, but it is
of course possible to adjoin more than one adjunct tree to the same
node of the target tree. The adjunction operation consists in merging
the root of the adjunct tree with the target node (which amounts to
unifying their feature structures) and adding the daughter of the
adjunct tree root as a new daughter to the target node. This can take
place in any position among the already existing daughters of that
target node.5 See Figure 20.7b for a general schematic illustration of
sister adjunction.
A complication arises when an operator or periphery element targeting a
specific layer occurs between elements that are part of a different layer.
Examples are given in (4):
(4) a. He ate the apple completely.

b. John did not eat the apple.
In (4a), the nucleus, which consists of the verb ate, is followed by an argu-
ment, which is part of the core, and after that comes an adverb that is in the
nuclear periphery. In (4b), the tense operator did is placed between an
argument (part of the core) and the verb (also part of the core). But it should
attach at the clausal level. For the moment, we will ignore this complication
and come back to it in Sections 20.5 and 20.7.
20.3.3 Wrapping Substitution

Control constructions and extraction from complements pose a problem for
the modes of composition presented so far. Consider the examples of wh-
extraction in (5).
(5) a. What does John think Kim smashed?

b. What does John think Mary claimed Kim smashed?
Clearly, it would not be appropriate to assume a separate complex tem-

plate for each of these constructions. The syntactic representations are to
be composed of basic argument structure templates in a systematic way.
There are several options for achieving this goal, depending on the pre-
sumed inventory of elementary templates. First we need to decide on the
proper syntactic representations of the examples in (5). While sentences of
this type are discussed in the context of island constraints in Van Valin and
LaPolla (1997: 615) and Van Valin (2005: 273), no structural analysis is
provided there. Due to the nature of the embedding constructions, the
basic binary pre-core slot pattern [CLAUSE[PrCS . . .][CORE . . .]] shown in
Figure 20.1 does not apply to the present case. Figure 20.9 shows two
possible alternatives. The analysis in Figure 20.9a assumes a pre-core vari-
ant of the clausal subordination pattern [CLAUSE [CORE . . .][CLAUSE . . .]]. The

(a) (b)
SENTENCE SENTENCE
CLAUSE CLAUSE
PrCS CLAUSE
PrCS CORE CLAUSE
RP CORE CLAUSE
RP RP NUC CORE
RP NUC CORE
RP NUC RP NUC
what does John think Kim smashed what does John think Kim smashed
Figure 20.9 Two possible syntactic representations of wh-extraction from complements
(a) (b)
SENTENCE SENTENCE
CLAUSE CLAUSE
CLAUSE CLAUSE
PrCS CORE CLAUSE PrCS CLAUSE
RP RP NUC RP CORE
CLAUSE
CLAUSE
think RP NUC
CORE CLAUSE
CORE CLAUSE
smash
RP NUC
RP NUC
CLAUSE
CLAUSE think
claim
CORE CLAUSE
CORE
RP NUC
RP NUC
claim
smash
Figure 20.10 Wh-extraction via simple substitution (a) and wrapping substitution (b)
structure in Figure 20.9b, by contrast, assumes an additional clause node,

and the pre-core slot pattern is [CLAUSE[PrCS . . .][CLAUSE . . .]]. In the following,
we restrict the discussion to the first option since we regard it as difficult
to come up with an independent motivation for the additional clausal
node in Figure 20.9b.
Figure 20.10 sketches two ways of composing the syntactic representation
of example (5b). Figure 20.10a employs substitution only, but at the price of
assuming a special elementary template associated with think that has a pre-
core slot in addition to its normal argument slots. Assuming such a template
would raise the further problem of providing information about which of
the arguments within the embedded clauses is referred to by the referent
phrase in the pre-core slot. The templates in Figure 20.10b are more straight-
forward in this respect since they represent proper argument structure
templates in that the pre-core slot is locally connected to the core from
which the wh-word is extracted. The dashed edge between the CLAUSE node
dominating the pre-core slot and the lower CLAUSE node stands for a
dominance relation, that is, a (possibly empty) sequence of immediate

Figure 20.11 Wrapping substitution
dominance edges in the final derived tree. We call such an edge a d-edge.
Under this analysis, the long-distance dependency comes about by the
insertion of material at the d-edge, that is, between the pre-core slot and
the corresponding core.6 More flexible templates and a more complicated
tree composition mechanism are needed in this case. The clause node of the
smash structure is split in two and its upper part is unified with the upper
clause node of the think structure (similar to the adjunction mechanism
introduced in the previous section), thereby keeping the pre-core slot on the
left side of the structure; the lower part of the split node is substituted at the
lower clause node of the claim structure. The mode of composition just
described is referred to as wrapping substitution (Kallmeyer et al. 2013; Oss-
wald and Kallmeyer 2018).
Wrapping substitution has the general form depicted in Figure 20.11.7 By
definition, this operation involves a tree with a d-edge. Such an edge stands
for a dominance relation in the final derived tree, that is, it specifies a
place in an elementary tree where additional nodes and edges from other
elementary trees can be inserted.8 Wrapping consists basically of splitting
the tree at the d-edge and wrapping it around a target tree. More specific-
ally, with reference to Figure 20.11, the subtree β rooted at the lower node
(labelled X) of the d-edge fills a substitution slot in the target tree α while
the upper node of the d-edge (labelled Y) merges with the root of α.
Concerning this upper node, all descendants to the left (resp. right) of
the d-edge are located to the left (resp. right) of the target tree α in the
resulting combined tree, and all nodes of γ dominating the upper node
(labelled Y) of the d-edge also dominate the merged node after the wrap-
ping. The example in Figure 20.10b is a case of wrapping with empty γ and
δR, (i.e. a wrapping where the upper part only adds nodes to the left of the
d-edge but neither to its right nor above).9
It is instructive to compare the different tree composition options of
Figure 20.10 with respect to their applicability for linking, especially with
respect to template selection. As already explained, the syntactic inven-
tory does not provide a single template for the complex syntactic struc-
ture of (5b). Rather, the structure has to be composed from argument
structure templates which in turn are selected by the chosen lexical
entries. The composition in Figure 20.10a has the disadvantages that
the template selected by think is not an argument structure template

but has to be stipulated for embedded wh-questions, and that an

advanced mechanism for coreference is needed. The composition scheme
in Figure 20.10b, by comparison, has the advantage that the pre-core slot
can be immediately linked when the template is selected. Here, the
underlying assumption is that the syntactic inventory provides argument
structure templates for wh-fronting. Note that this assumption differs
slightly from RRG’s standard understanding of syntactic templates,
which would assume a reduced core template plus a pre-core slot tem-
plate in this case (Van Valin 2005: 15). Elementary trees in LTAG, by
comparison, are commonly assumed to have substitution nodes for all
arguments of the lexical head, irrespective whether they are realized
within the CORE or outside of it (cf. Section 20.3.4). A further decom-
position of elementary trees can be expressed by means of tree descrip-
tions in the meta-grammar (cf. Section 20.4.2).
20.3.4 Extended Domain of Locality

As mentioned in Section 20.3.1, an important characteristic of the LTAG
formalism is the extended domain of locality (EDL) of elementary trees (Joshi
and Schabes 1997: 95f ), which means that elementary trees represent full
argument projections and that they can have a complex constituent struc-
ture. As explained above, the present formalization of RRG builds on a
similar assumption. This is particularly crucial in the case of long-distance
dependencies across clausal complements, as in Figure 20.10b. Here, the
wh-element in the pre-core slot originates from the same elementary tree as
the verb smash that it depends on. In the final derived structure, the
two are separated from each other by the intervening structures of claim
and think.
A possible alternative solution, avoiding this extended domain of locality
and using only simple substitution, is to have the information about the wh-
marked argument percolate to the top of the tree. A constraint-based for-
malization of this percolation process could roughly work as follows. The
core node of the reduced template for smash carries a (set-valued) feature
which contains the referential index of the participant not locally realized
as well as its wh-marking. A general constraint then ensures that clause and
sentence nodes collect the non-realized indices of their subordinate clauses
and cores minus the indices that are realized in pre-core slot daughters and
the like. This way of bookkeeping for modelling long-distance dependencies
is in fact closely related to the use of ‘slash’ or ‘gap’ features in the
approaches of Sag and Wasow (1999) and Ginzburg and Sag (2001),
among others.
To sum up, the approach just discussed can get along with simple
substitution at the price of a considerable amount of bookkeeping. By
comparison, the approach exemplified in Figure 20.10b allows a fully local
argument linking but requires a more complex method of tree

composition. The former strategy is employed in GPSG/HPSG-related frame-

works while the latter strategy is characteristic of approaches in the line of
LTAG (cf. Kroch 1987).
20.3.5 Formal and Computational Properties of Tree

Wrapping Grammars
The type of tree rewriting grammar formalism introduced above, with
composition operations sister adjunction, substitution and wrapping substi-
tution, is called Tree Wrapping Grammar (TWG, Kallmeyer et al. 2013; Kall-
meyer 2016). Note that Kallmeyer et al. (2013) use a slightly different
definition of wrapping substitution, which, however, gives rise to more
binary trees. We use the TWG version presented in Kallmeyer (2016) and
Osswald and Kallmeyer (2018).
One can investigate the formal properties of TWGs independently of the
concrete shape of elementary trees chosen for RRG. We will not go into
detail here but restrict ourselves to pointing out the following (see Kall-
meyer 2016). TWGs are more powerful than context-free grammars (CFGs),
which is due to the wrapping substitution operation. The expressive power
depends in particular on how many d-edges are allowed to stretch across a
single node in the final derived tree. In our example in Figure 20.10b, for
instance, we have one single d-edge stretching across the roots and cores of
the claim and the think elementary trees (see the grey dashed edge in
Figure 20.12, which is the d-edge from the elementary tree of smash).
A TWG where the maximum number of d-edges stretching across a node is
limited to some constant k is called a k-TWG, and Kallmeyer (2016) links
k-TWG to simple Context-Free Tree Grammars (CFTG, Kanazawa 2016) of
CLAUSE
PrCS CLAUSE
RP CORE
RP NUC CLAUSE
think CORE
RP NUC CLAUSE
claim CORE
RP NUC
smash
Figure 20.12 D-edge in final derived tree for Figure 20.10b

rank k by showing that for every k-TWG, a simple CFTG of rank k can be
constructed that generates the same language. Simple CFTGs of rank k are,
in turn, equivalent to well-nested Linear Context-Free Rewriting Systems
(LCFRS) of fan-out k þ 1. Consequently, k-TWGs are in particular mildly
context-sensitive.
The notion of mildly context-sensitive languages and formalisms was
introduced in Joshi (1985) in an attempt to characterize the amount of
context-sensitivity required for natural languages, where context-sensitivity
is used in a formal language sense, in contrast to context-free languages and
grammars. Roughly, a mildly context-sensitive grammar formalism is able
to generate more than all context-free languages, can account for cross-
serial dependencies, is polynomially parsable,10 and has the constant-
growth property, which means that, if we order the sentences of a language
according to their length, the length grows in a linear way. It is interesting
that coming from RRG, a typologically motivated linguistic theory, and
developing a tree rewriting formalization of it, we end up within a class of
formalisms that seems to support Joshi’s hypothesis that natural languages
are mildly context-sensitive.
20.3.6 Crossing Branches and Discontinuous Structures

The formal syntactic framework introduced so far presupposes ordered trees
(cf. Section 20.3.1). This means that every two nodes stand either in a
dominance relation or a linear precedence relation, and if one node pre-
cedes another node then all descendants of the first node precede the
descendants of the second node. It follows that crossing branches, which
are a frequent phenomenon in RRG syntactic analyses, are not supported by
the formalism. Crossing branches often occur with peripheral adjuncts at
the nucleus or the clause, for example. Because of the single syntactic tree
assumption of the present approach (cf. Section 20.2.2), operators are
another potential source of crossing branches. The proposed solution in
both cases is to attach the respective elements at a lower node of the phrasal
skeleton, in order to avoid a crossing of branches in the tree, while keeping
track of their scope by means of appropriate features. Details on how to do
this will be given in Sections 20.5 and 20.7.
Another reason for crossing branches can be discontinuities in the con-
stituents in the clausal skeleton. Discontinuities in the nucleus may occur
with predicates that are multi-word expressions (e.g. particle verbs) or com-
plex predicates (nuclear cosubordinations). An example of the latter is given
by the English resultative construction shown on the left of Figure 20.13. In
such cases, we assume that the discontinuous node can be split into two
components, both carrying the same identifier nuc-id as a feature (see the
right of Figure 20.13), in order to signal that the two nodes are actually
components of the same nucleus.11 Note that the two components would be
part of the same elementary tree.

SENTENCE SENTENCE
CLAUSE CLAUSE
CORE CORE
NUC NUC[ 1] NUC[ 1]
RP NUC RP NUC RP NUC RP NUC
V[ +] AP[ +] V[ +] AP[ +]
John painted the house red John painted the house red
Figure 20.13 Discontinuous complex predicates
20.4 Specifying the Syntactic Inventory
20.4.1 The Structure of Syntactic Templates

The syntactic templates are the basic building blocks of syntactic represen-
tations. Let us now turn to the question of how to specify the templates of
the syntactic inventory in a systematic way. The discussion of the two
approaches to wh-fronting in Section 20.3.3 shows that the question of which
templates must be available in the inventory is not independent of the
modes of composition employed. If extraction structures are fully linked
locally, as in Figure 20.10b, then there is no need for a pre-core slot template
to be attached separately. As mentioned in Section 20.3.1, the present
formalization proposal of RRG syntax assumes that argument structure
templates have slots for each of the arguments of the lexical anchor. We
call this the full argument projection assumption.12
For example, the templates in Figure 20.14 represent (some of the) alter-
native realization patterns of transitive verbs in English. They would be used
for the different forms of eat in (6) (in the respective order in which they
occur in the figure).
(6) a. The boys ate the entire cake.

b. The cake was already eaten.
c. The entire cake was eaten by the boys.
d. What did the boys eat today?
e. I planned to serve the cake that the boys just ate.
f. I expected them to eat the cake.
Notice the two templates for the passive (b and c), one in which only the
undergoer occurs as an argument, and a second template which includes
the realization of the actor by a peripheral by-phrase.13 Standard RRG would
probably regard the peripheral by-phrase as a separate template that can be
adjoined to the passive core. In the context of the present framework, the

(a) (b) (c)

CORE CORE CORE
RP NUC[VOICE act] RP RP NUC[VOICE pass] RP NUC[VOICE pass] PP [PERI +]
V[PRED +] AUX V[PRED +] AUX V[PRED +] P RP
by
(d) (e) (f)
CLAUSE NUC∗N CORE
PrCS CORE CLAUSE[PERI +] NUC[VOICE act] RP
RP[WH +] RP NUC[VOICE act] PrCS CORE V[PRED +]
V[PRED +] RP[REL +] RP NUC[VOICE act]
V[PRED +]
Figure 20.14 Basic transitive predication template for English with variants
question is whether the by-phrase should be added by (sister) adjunction like

an adverbial. By listing the two realization patterns for the passive as
elementary templates in Figure 20.14, we decide against adjunction in this
case. The main disadvantage of modelling the by-phrase by adjunction
would lie in the constraints that need to be imposed, since adding a periph-
eral by-phrase that encodes the actor is restricted to specific constructional
circumstances. The optional peripheral by-phrase is part of the description
of how passive voice is realized in English. Of course, a mere enumeration of
the two passive templates in Figure 20.14 is not satisfactory for a theory of
grammar. At some point, the theory should state explicitly that it is the
addition of the by-phrase which relates the second template to the first. In
our framework, this relation is encoded at the level of template specifications
(cf. Section 20.4.2).
Template (d) in Figure 20.14 represents an elementary pre-core slot
argument structure template. This differs from the template system
informally suggested in Van Valin (2005: 15), in which the structure in
question is composed of a reduced core template and a pre-core slot
clause fragment (cf. Figure 20.1b). Again, we propose that compositions
of this type are best modelled at the level of template specifications.
Template (e) is used when the predicate is the anchor of a restrictive
relative clause where the relative pronoun is the object. In this case, the
peripheral tree is an adjunct tree that is intended to be adjoined to the
nucleus of the noun phrase that it modifies. The last template (f ) is the
one for a subjectless infinitival core, which can be embedded as an
argument of category CORE, as in (6f).

20.4.2 Template Specification by Tree Descriptions

In the previous section, we raised the question of how to characterize the
(universal and language-specific) syntactic templates in a systematic way.
The templates proposed above as elementary are more complex than the
fragmentary templates originally assumed in the RRG literature. The key
advantage of the formalization approach described here is that the com-
position of syntactic representations can be reduced to the three modes
of composition introduced in Section 20.3, namely simple substitution,
sister adjunction and wrapping substitution. Furthermore, linking can be
performed locally within elementary trees (cf. Section 20.8). However, this
leaves us with the problem of how to describe in which way the tem-
plates are built from more elementary components, and how they are
related to each other. As mentioned before, it would be rather unsatis-
factory if we had to regard the two passive templates in Figure 20.14 as
independent units without being able to make explicit the relation
between them.
The proposed solution is to treat syntactic templates as minimal models of
tree descriptions. Relations between templates can then be captured by
relating the respective descriptions. For instance, the specification of
the passive template with by-PP consists of the specification of the simple
passive template plus a specification of the by-PP and constraints on its
position. The use of tree descriptions for specifying syntactic templates in
a modular way is inspired by the meta-grammar approach of Crabbé and
Duchier (2005), where a meta-grammar is basically a system of tree descrip-
tions that defines the syntactic inventory as the set of the associated
minimal models.
Tree descriptions in this sense consist of dominance and precedence
constraints as well as category and feature assignments. Consider the
example specifications in Figure 20.15, which are depicted in tree-like
core-spine core-clause pre-core-slot pre-nuc-rp post-nuc-rp

CORE CLAUSE CLAUSE CORE CORE
NUC CORE PrCS ≺ CORE RP ≺ NUC NUC ≺ RP
V[PRED +]
clause-spine:= base-transitive :=
core-spine ∧ core-clause clause-spine ∧ pre-nuc-rp ∧ post-nuc-rp
CLAUSE CLAUSE
CORE CORE
NUC RP ≺ NUC ≺ RP
V[PRED +] V[PRED +]
Figure 20.15 Example specifications of syntactic fragments

diagrams but are to be read as tree descriptions. For instance, the specifica-
tion with the name pre-core-slot says that there are three distinct nodes n0, n1
and n2, labelled respectively by CLAUSE, PrCS and CORE, where n1 and n2 are
daughters of n0 and n1 immediately precedes n2 (expressed by ≺).
Figure 20.15 illustrates how the basic transitive template of Figure 20.14
can be defined by a piecewise combination of such specifications. The pre-
core slot template of Figure 20.14 can likewise be defined by conjoining
the specifications clause-spine, pre-core-slot, and pre-nuc-rp. In this way, common
components of elementary templates are made accessible in the meta-
grammar.14 Furthermore, syntactic tree descriptions can be linked to
descriptions of semantic representations (either predicate-logical formulas
or frames), and the meta-grammar can include a constraint-based formula-
tion of RRG’s linking algorithm (cf. Kallmeyer et al. 2016). The syntax–
semantics interface will be briefly discussed in Section 20.8.
20.5 Formalizing the Operator Projection
20.5.1 Operators in RRG

RRG links the representation of operators to the layered structure of the
clause. Recall that the layered structure reflects the distinction between
predicates, arguments and non-arguments. The core layer consists of the
nucleus, which specifies the verb, and its arguments. The clause layer
contains the core as well as extracted arguments. Operators are closed-class
grammatical categories such as aspect, modality and tense. Each type of
operator is assumed to attach to a specific layer: for instance, tense operators
attach to the clause, modality to the core, aspect to the nucleus (see
Table 20.1 for the mapping between operator types and corresponding
layers). Moreover, the surface order of the operators reflects their attach-
ment site in that the higher the layer an operator is attached to, the further
Table 20.1 Operators in the layered structure of the clause

(cf. Van Valin 2005: 9)
Layer Operators
Nucleus Aspect (ASP )

Negation (NEG )
Directionals (event orientation)
Core Directionals (participant orientation)

Modality (MOD ) (ability, permission, obligation, etc.)
Negation (NEG )
Clause Status (STA ) (epistemic modals, external negation)

Tense (TNS )
Evidentials
Illocutionary Force (IF )

away from the nucleus the operator occurs on the surface. The mapping
from operators to levels of the layered structure explains (i) the scope
behaviour of operators, since structurally higher operators take scope over
lower ones, and (ii) surface order constraints for operators; higher operators
are further away from the nucleus of the structure.
The problem is that the constituent and the operator structure are not
completely parallel; one can have structures where an operator belonging to
a specific layer is, on the surface level, surrounded by elements belonging to
a lower layer in the constituent structure. Examples of a clause-level tense
operator occurring within the core are given in (7) and (8) (taken from Van
Valin, 2005: 10) for English and Turkish, respectively.
(7) [Mary enter-edTNS the room]CORE.
(8) [Gel- emi- yebil- ir- im]CORE.

come- able.neg MOD - psbl STA - aor TNS - 1sg
‘I may be unable to come.’
In (8), the clause-level status and tense operators occur between the verb and
the pronominal affix, which is part of the core.
Even though the constituent structure and the operator structure are not
fully aligned, they depend on each other. Their hierarchical order is the
same and the existence of a layer in the operator projection requires that
this layer also exists in the constituent structure. For instance, one can only
have clause-level operators if a clause node exists in the constituent
structure.
While the ordering among the operators is thus systematically correlated
with the scope given by their attachment site at the clausal skeleton, the
surface order of the operators relative to arguments and adjuncts is much
less transparent and would require crossing branches if everything were
captured in a single tree with operators attaching where they take scope. For
this reason, RRG usually represents the constituent structure and the oper-
ator structure as different projections of the clause. The syntactic represen-
tation for (7) on the left side of Figure 20.16 illustrates this idea: the upper
part gives the constituent projection while the lower part gives the operator
projection. If we integrate the operator projection into the clausal skeleton
of the constituent structure, we obtain the tree with crossing branches
shown on the right of Figure 20.16, due to the above-mentioned mismatches
between the two projections.15
Besides crossing branches, additional complications arise from the
following two facts: first, a functional element can contribute more than
one operator and, second, an operator can be distributed over more than
one element in the sentence. An example is (9) where the functional element
will contributes both tense and illocutionary force (IF), while aspect (ASP) is
jointly contributed by be and -ing.
(9) WillIF,TNS they beASP leav-ingASP?

CLAUSE CLAUSE
CORE
CORE
RP NUC RP
V NUC
RP RP
Mary enter-ed the room V TNS
V
Mary enter -ed the room
NUC
CORE
CLAUSE TNS
Figure 20.16 Constituent structure and operator projection for (7)
CLAUSE
CORE
CLAUSE
RP NUC
V
will they be leav -ing IF TNS CORE
V RP NUC
ASP NUC ASP ASP V ASP

CORE
will they be leav -ing
TNS CLAUSE
IF CLAUSE
RRG assumes the structure on the left of Figure 20.17 (adapted from Van
Valin 2005: 14). If the operator projection is integrated into the constituent
structure, we obtain the graph on the right of Figure 20.17, which is not
even a tree but only a directed acyclic graph since the node labelled will has
two incoming edges. Note, however, that as long as nodes with more than
one incoming edge arise only from contributing several operator categories
to the same level of the layered structure, the graph can easily be turned
into a tree by putting all these categories into a single node with a label (or,
rather, a feature structure) that captures the fact that the element is an
operator that contributes several categories to the same layer. The node
dominating will in Figure 20.17, for instance, could have a category OP with
features TNS ¼ fut and IF ¼ int.
Cases with a single element contributing two operators that take scope at
different layers exist as well. Examples are finite modal verbs as in (10) that
provide modality at the core level and tense at the clause.
(10) John mightTNS,MOD win.

CLAUSE
CORE
CLAUSE
RP NUC
V CORE
John might win
V RP NUC
NUC TNS MOD V

MOD CORE
John might win
TNS CLAUSE
The corresponding constituent structure and operator projection in RRG is

shown on the left of Figure 20.18, and on the right we see the graph that one
would obtain when integrating the operator projection into the constituent
tree. This example shows that if we integrate the operator projection into
the constituent tree while assuming that each operator has a separate node
that is the daughter of its syntactic layer, we obtain directed acyclic graphs
instead of trees, which might come with considerable complications con-
cerning the formal properties of such a framework. In particular, it is clear
that such structures cannot be generated with the TWGs introduced in the
previous section.
If we choose to disentangle the operator projection to a certain extent
from the constituent structure, we have two options. Either we can generate
the operator projection tree separated from but to a certain extent in
parallel with the constituent structure. This is what Johnson (1987) proposed
in his formalization, which uses two different context-free grammars, one
for analysing the sequence consisting of the verb plus arguments, and one
for the sequence consisting of the verb plus operators (cf. Section 20.2.1).
Each takes those parts of the sentence as input (or yield) that corresponds to
its set of terminals. This approach makes sure that verb and arguments, on
the one hand, and verb and operators, on the other, appear in the right
order, independent from each other. But there are some shortcomings: First,
the formalization does not guarantee that each layer targeted in the oper-
ator projection is actually present in the constituent structure. Secondly, a
problem remains with elements that contribute more than one operator,
possibly to more than one layer, since the different projection grammars are
each assumed to generate trees.
20.5.2 A Feature-Based Implementation of the Operator Projection

In the following, we will present the proposal of Kallmeyer and Osswald
(2017), which integrates operators according to their surface position into
the constituent structure, thereby avoiding crossing branches, while

keeping track of the operator projection within the feature structures. The
classification of an element as an operator and the contribution of that
element to the various layers is captured within the category of its node.
More specifically, an operator element has the category label OP and a
feature structure that specifies its contribution to the CLAUSE (feature cl),
the CORE (feature co) and the NUC (feature nuc) layer. The tree structures
corresponding to the sentences (9) and (10) are shown in Figure 20.19. Note
that the operator scope information is now entirely captured within the
feature structure at the OP node and does not depend on the category of the
node to which the OP attaches.
In order to obtain trees of the type given in Figure 20.19, we use sister
adjunction for adding operators. This leads to spurious ambiguities since
operators could in principle adjoin to any of the three layers that they
attach to, provided this does not yield crossing branches. In the first tree
in Figure 20.19 for instance, we could also attach the clausal operator will
lower, at the CORE node, and be could attach higher, also at the CORE
node. In the following we assume that among the different combinations
of attachment sites that are possible, the preferred one is the one where
operators are placed as close as possible to the layer that corresponds to
their scope (if it scopes at several layers, the lowest is considered
relevant).
Aside from this spurious ambiguity issue, if we adjoin operators in an
unrestricted way, it leads to undesired overgeneration since we can leave
CLAUSE
IF int
OP CL
TENSE fut
CORE
will RP NUC
they OP NUC ASP prog V OP NUC ASP prog
be leav -ing
CLAUSE
CORE
CL [TENSE pres]
RP OP CO [MOD epistemic]
NUC
John might V
win
Figure 20.19 Encoding the operator structure in node features for (9) and (10)

out operators that are obligatory, adjoin several operators of the same type
even in cases where this is ungrammatical, and we can generate ungram-
matical linear orders of operators, for instance the ones in (11).
(11) a. *They be will leaving?

b. *John not might win.
Kallmeyer and Osswald (2017) introduce edge features in order to express

the necessary constraints on adjunction. The basic idea is that each edge (or,
under a different but equivalent view, each node) should be able to pass
information on elements already attached and elements still required to the
left and to the right. Once the derivation is completed, the edge information
of sister nodes must be compatible with each other. To this end, each node
(or, as Kallmeyer and Osswald (2017) put it, each edge) has a left and a right
feature structure that is not part of the proper feature structure of the node
but interacts with the left and right feature structures of neighbouring
nodes. They are graphically depicted in shaded boxes to the left and the
right of the node in question (see Figure 20.20). (The feature structure of the
node itself will often be omitted or, if needed, given in the middle.) Note
that the features nuc, co and cl mentioned above and exemplified in
Figure 20.19 are not part of the edge features but are proper node features.
Edge features are less about the actual content of nodes but rather about
requirements and information that need to be passed around in order to
constrain syntactic composition.
Let us illustrate this approach by using it to characterize a certain
operator as obligatory. In Figure 20.20, we have a verb without tense
marking. Therefore, to the left of the NUC node, we signal [tns ]. The
features we are using here are mostly binary features (i.e. values are only þ
or ), and their intended meaning is that a value expresses the absence of
something while a value þ expresses the presence of or the requirement
(presence in the derived tree) for something. The feature [tns þ] on the right
CLAUSE
PrCS TNS + CORE
RPwh TNS 3 RP TNS 3 TNS − NUC
what CLAUSE∗ the girl V
TNS + OPtns TNS − say
does
Figure 20.20 Obligatory adjunction of a tense operator

2 [: : : ] • • 3 [: : : ]
• 1 [: : : ] 1 [: : : ] • 2 [: : : ] • • 3 [: : : ]
unification between feature percolation feature percolation

neighbouring edges on the left fringe on the right fringe
Figure 20.21 Final feature unifications
of the PrCS signals that to the right of this node, a tense operator has to be
present. If no tense operator adjoins, the absence and the requirement of the
presence will eventually unify (we will explain later why) and this will lead
to a unification failure. If, however, a tense operator adjoins as in
Figure 20.20, the requirement is satisfied. The features to the left and to
the right of the tense operator tell us that to its right, no other tense
operator exists while to its left one can pass the information that there is
now a tense operator. (OPtns is short for category OP with feature structure
[cl [tense . . .]].)
Let us explain now exactly how edge features work (see Kallmeyer and
Osswald 2017), in particular how they percolate through the tree (cf.
Figure 20.21). As mentioned, nodes can have special left and right feature
structures. In the final derived tree, the left feature structure of a node ν
unifies with the right feature structure of its immediate sister to the left.
Furthermore, the left feature structure of a node ν that does not have a sister
to the left unifies with the left feature structure of the mother of ν, provided
this mother is not the root node of an elementary tree or the lower node of a
d-edge. Similarly, the right feature structure of a node ν that does not have a
sister to the right unifies with the right feature structure of the mother of ν,
again provided this node is not the root node of an elementary tree or the
lower node of a d-edge. In our example, once we have performed the
adjunction, we obtain the tree at the top of Figure 20.22, and after the final
edge feature unifications, the result is the tree at the bottom. Such edge
features can be used to require certain adjunctions and, as is the case in this
example, to require them exactly once.
The adjunction of the operators needs to be controlled in the following two
respects: (i) Adjoining an operator is obligatory if the information conveyed by
the operator is required for a sentence to be complete. (ii) The scope-related
ordering of the operators must be respected. In our approach, these constraints
are both implemented with the help of edge feature structures; we have just
seen an example of type (i). We will now introduce features that guarantee (ii)
as well. Concretely, we want to achieve that, from the nuclear predicate
outwards, one encounters first the NUC operators, then the CORE operators
and then the CLAUSE operators. To this end, we assume an edge feature ops
(for operator structure) that specifies which operator projection layer(s) have

before final unifications:

CLAUSE
PrCS TNS + TNS + OPtns TNS − CORE
RPwh does TNS 3 RP TNS 3 TNS − NUC
what the girl V
say
after final unifications:

CLAUSE
PrCS TNS + TNS + OPtns TNS − TNS − CORE
RPwh does TNS − RP TNS − TNS − NUC
what the girl V
say
Figure 20.22 Obligatory adjunction of a tense operator: derived tree
been reached so far. Its value is a feature structure with features cl_l, co_l
and nuc_l for the three layers, each with possible values þ or . These
features are used in such a way as to guarantee that nuclear, core and
clausal operators have to appear in this order when moving outwards in
the sentence, starting from the nuclear predicate.
Figure 20.23 illustrates this mechanism by applying it to the operator
structure of the sentence in (12).
(12) John might not win.
For instance, a core operator such as not that adjoins to the left of the
predicate has a requirement to the right that the level cl_l have a value
(i.e. has not been reached yet). To the left, it just gives the information
co_l þ, which has the effect of disallowing nuclear operators to appear
here. The operator might, which contributes not only tense at the clausal
level but also modality at the core level, comes with the same require-
ments to its right. But, in contrast to not, it states to the left of the edge
that now cl_l has value þ (as well as co_l). An inverse order (John not
might win) is therefore excluded. To the left of might, there cannot be any
further core (or nuclear) operators. The derived tree for sentence (10) is
shown in Figure 20.24.
Recall that in addition to the ops feature, nodes of category OP have
features nuc, co and cl that indicate the contribution of the operator.
For example, since might contributes tense at the clause level and epistemic
modality at the core level, its OP feature structure is [cl[tense pres], co[mod
epistemic]] (cf. Figure 20.19). (We left this out of Figures 20.23 and 20.24 for

CLAUSE
TNS + CORE
TNS 3 TNS 3
OPS 4
RP OPS 4
NUC
⎡ ⎤ ⎡ ⎤
TNS −
⎡ ⎤ TNS −
⎡ ⎤
⎢ NUC_L − ⎥ ⎢ NUC_L − ⎥
John ⎢ ⎥ V ⎢ ⎥
⎣OPS ⎣CO_L − ⎦⎦ ⎣OPS ⎣CO_L − ⎦⎦
CORE ∗ CL_L − CL_L −
⎡ ⎤
TNS + win
⎣ TNS −
CL_L + ⎦ OPmod,tns
OPS OPS CL_L −
CO_L +
might
CORE∗
TNS − TNS −
OPS [CO_L +]
OPneg OPS CL_L −
not
Figure 20.23 Keeping track of the operator projection in edge features
the sake of readability.) Since the features of the OP nodes specify which
operator projection layer(s) the operator belongs to, we can deterministic-
ally map a derived tree to the standard RRG structure in which the constitu-
ent structure and the operator projection are separated.
20.6 The Construction of Complex Sentences
20.6.1 Coordinate, Subordinate and Cosubordinate Constructions

A crucial assumption of RRG concerning the structure of complex
sentences is the distinction between embedded and non-embedded
dependent structures. Embedded dependent structures correspond to
subordinations. By contrast, non-embedded dependent structures, which
are referred to as cosubordination structures, have the basic form
[[ ]X [ ]X]X. It is characteristic of this type of construction that operators
that apply to category X are realized only once but have scope over both
constituents. Cosubordination differs from the coordination of two inde-
pendent structures in that the latter type of construction has the form
[[ ]X [ ]X]Y, where Y is a category one level above X in the layered structure.
The general schemas for non-subordinate structures are shown in

CLAUSE
TNS + CORE
⎡ ⎤
TNS +
TNS 3 TNS 3 ⎣ TNS − TNS − TNS −
RP CL_L + ⎦ OPmod,tns OPneg NUC
OPS 4 OPS 4 OPS OPS CL_L − OPS [CO_L +] OPS CL_L −
CO_L +
⎡ ⎤ ⎡ ⎤
TNS −
⎡ ⎤ TNS −
⎡ ⎤
⎢ NUC_L − ⎥ ⎢ NUC_L − ⎥
John might not ⎢ ⎥ ⎢ ⎥
⎣OPS ⎣CO_L − ⎦⎦ V ⎣OPS ⎣CO_L − ⎦⎦
CL_L − CL_L −
CLAUSE
⎡ ⎤
win
⎡ ⎤ TNS −
TNS + ⎡ ⎤
⎢ CL_L − ⎥
⎣ CL_L + ⎦ CORE ⎢ ⎥
OPS ⎣OPS ⎣CO_L − ⎦⎦
CO_L +
NUC_L −
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ TNS − TNS − TNS −
TNS + TNS + TNS + TNS − TNS − ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
⎢ NUC_L − ⎥ ⎢ NUC_L − ⎥ ⎢ NUC_L − ⎥
⎣ CL_L + ⎦ RP ⎣ CL_L + ⎦ ⎣ CL_L + ⎦ OPmod,tns ⎣ CL_L − ⎦ ⎣ CL_L − ⎦ OPneg ⎢
⎣OPS ⎣CO_L − ⎦⎦
⎥ ⎢ ⎥
⎣OPS ⎣CO_L − ⎦⎦ NUC
⎢
⎣OPS ⎣CO_L − ⎦⎦
⎥
OPS OPS OPS OPS OPS
CO_L + CO_L + CO_L + CO_L + CO_L +
CL_L − CL_L − CL_L −
⎡ ⎤ ⎡ ⎤
TNS −
⎡ ⎤ TNS −
⎡ ⎤
⎢ NUC_L − ⎥⎥ ⎢ NUC_L − ⎥
John might not ⎢ ⎢ ⎥
⎣OPS ⎣CO_L − ⎦⎦ V ⎣OPS ⎣CO_L − ⎦⎦
CL_L − CL_L −
win
Figure 20.24 Derived tree for (10) before (top) and after (bottom) final edge feature unification
X Y
… …
X X … X X X … X
Figure 20.25 RRG template schemas for non-subordinate nexus types
sister adjunction
CLAUSE
CORE* CORE* CORE
CORE CORE CORE

CLAUSE
NUC NUC NUC
CORE
CLAUSE CORE CORE CORE
CORE CORE CORE NUC NUC NUC
CORE CORE CORE CORE CORE
NUC NUC NUC

wrapping substitution
Figure 20.26 Two ways of compositionally deriving cosubordination constructions
Figure 20.25 (cf. Van Valin and LaPolla 1997: 507 and Van Valin 2005: 224).
However, these structures cannot serve as elementary trees in the sense
introduced in Section 20.3 since there are infinitely many of them; that
is, they have to be derived by operations in the syntax.
Among the tree composition operations described in Section 20.3, both
sister adjunction and wrapping substitution allow the derivation of cosu-
bordinate structures, as sketched in Figure 20.26 for the case of core
cosubordination. The following two examples of multi-verb constructions
show possible use cases for the two kinds of derivations. Both examples
describe transitive motion scenarios. The example in (13), taken from
Ullrich (2011), illustrates a common pattern for simultaneous event con-
structions in Lakhota (Siouan) (cf. Ullrich 2018; Osswald and Van Valin
2022).
(13) Yu-slóhaŋ a-wíčha-Ø-ye.

by.pulling-slide am-3pl.ug.anim-3sg.ac-go
‘She was dragging them away.’
In Lakhota, arguments are marked by pronominal affixes on the main verb

at the end of the clause. The simultaneous event construction in (13) con-
tains a second verb that precedes the main verb and is dependent in that it is
morphophonologically reduced and does not carry any pronominal

CLAUSE
CORE[COSUB +]∗
COSUB + CORE[COSUB +]
COSUB + CORE
COSUB − CORE
NUC
NUC
V
V
Figure 20.27 Core cosubordination via sister adjunction: analysis of (13)
markings. The construction has the properties of a core cosubordination

(Ullrich 2018). Since there are no shared arguments to be taken care of
syntactically, due to the head marking, and because the construction has
the same-subject/same-object constraint, the cosubordinate structure can be
derived most naturally by sister adjunction, as illustrated in Figure 20.27.
Under this analysis, the main verb selects an elementary tree that already
has two core nodes and requires the adjunction of an additional core
daughter. This requirement can be captured via a binary edge feature cosub
which signals that a structure is a cosubordination structure. In addition,
the higher CORE node of a CORE cosubordination can have a feature
[cosub þ] while the lower ones are marked [cosub ], which ensures that
the adjunction takes place at the upper CORE node. Note that these node
features do not interact with the edge features.
The Japanese example in (14) (from Croft et al. 2010: 219) shows the use
of the te- construction for combining cause, manner and direction of
motion.
(14) Watashi wa taru o korogashi-te chikashitsu ni ire-ta.

1sg top barrel acc roll(caus)-te basement loc take.into-pst
‘I rolled the barrel into the basement.’
Assuming that (14) is also an instance of a core cosubordination construc-

tion, the derivation seems best described as an application of wrapping
substitution (see Figure 20.28). Under this analysis, the first two argu-
ments are introduced by the first verb and are shared by the second verb.
The elementary tree chosen for the second verb has, therefore, a reduced
number of arguments. Its remaining arguments are ‘controlled’ by the
elementary tree of the te -marked first verb, like in a control construction.
Multiply embedded control and matrix-coding constructions pose a
number of challenges for RRG’s syntactic analysis as presented in Van Valin
and LaPolla (1997) and Van Valin (2005). Consider the examples in (15) and
their syntactic representations in Figure 20.29.

(a)
CLAUSE
CORE
CORE CORE
RP RP NUC RP NUC
watashi wa taru o korogashi-te chikashitsu ni ire-ta
(b)
CLAUSE
CORE [COSUB +] CORE [COSUB +]
CORE CORE CORE
RP RP NUC RP NUC
korogashi-te ire-
Figure 20.28 Core cosubordination analysis of (14) derived by wrapping substitution
SENTENCE
CLAUSE
CORE CORE CORE
RP NUC RP LM NUC RP LM NUC RP

SENTENCE
Mary expected John to ask Kim to clean the floor
CLAUSE
CORE
CORE
SENTENCE CORE CORE
CLAUSE
RP NUC LM NUC RP LM NUC RP
CORE
CORE John tried to persuade Kim to clean the floor
CORE CORE
RP NUC RP LM NUC LM NUC RP
John told Kim to try to clean the floor
Figure 20.29 Syntactic representations of the examples in (15)
(15) a. Mary expected John to ask Kim to clean the floor.

b. John tried to persuade Kim to clean the floor.
c. John told Kim to try to clean the floor.
Structures like the ones in Figure 20.29 show a mismatch between syntax
and semantics in the following sense: semantically, the infinitival

(a)
CORE CORE CORE CORE CORE CORE CORE CORE
RP NUC RP NUC RP NUC RP RP NUC RP NUC RP NUC RP
expected (to) ask (to) clean
(b)
CLAUSE CLAUSE
CLAUSE
CORE CORE CORE
CORE CORE CORE CORE CORE CORE
RP NUC NUC RP RP NUC NUC RP
tried (to) persuade
Figure 20.30 Composition of templates by wrapping substitution for (15a) and (15b)
CORE CORE PrCS CORE CORE CORE PrCS CORE CORE CORE
RP NUC RP RP NUC NUC RP RP RP NUC RP NUC NUC RP
expected (to) ask (to) clean
Figure 20.31 Composition of the syntactic representation of (16a)
complements are arguments of the respective control predicates, but syn-

tactically, they do not behave like core arguments. Example (15a) is instruct-
ive in showing how multiply embedded constructions can give rise to
coordinating syntactic configurations. Simple substitution cannot be
applied here if we assume that the verbs expect and ask select elementary
templates that contain a core daughter of the clause for the infinitival
complement. Figure 20.30a sketches a possible solution that uses wrapping
substitution, where split nodes carry differing top and bottom categories.
Figure 20.30b shows that the same compositional mechanism works for
control verbs such as try. which call for a core cosubordination template.
Moreover, the treatment of wh-fronting introduced above straightforwardly
extends to embedded control and matrix-coding constructions like those
in (16).
(16) a. Whom did Mary expect John to ask to clean the floor?
b. What did Mary expect John to ask Kim to clean?
For example, the syntactic representation of (16a) can be composed of

elementary argument structure templates as illustrated in Figure 20.31.
Note that clausal complements as in (17a) also show a mismatch between
syntax and semantics in that they do not attach to the core but to the clause.

(a)
CLAUSE CLAUSE
CORE CLAUSE CORE
RP NUC RP NUC RP
John thinks Kim smashed the vase
(b) CLAUSE CLAUSE
PrCS CLAUSE OPtns CORE CLAUSE
RPwh CORE does RP NUC
what RP NUC John think
Kim smashed
Figure 20.32 Subordination via simple and wrapping substitution for (17a) and (17b), respectively
But in contrast to the cosubordination cases just discussed, multiple embed-

dings of clausal complements correspond to embeddings on the syntactic
side, though at the clause level and not at the core level. For this reason,
simple substitution is sufficient for clausal complementation (cf.
Figure 20.32a).
(17) a. John thinks (that) Kim smashed the vase.

b. What does John think (that) Kim smashed?
However, as already discussed in Section 20.3.3, cases of extraction out of

subordinated constituents, like in (17b), require wrapping substitution
(cf. Figure 20.32b).
20.6.2 Operators in Complex Sentences

Operators in cosubordination constructions that appear in one of the com-
ponents and that target the cosubordination layer have scope over all
components. In the following, we will illustrate how this characteristic
property of cosubordination can be formally achieved, going through
one example involving sister adjunction and one example involving
substitution.
The Turkish sentence in (18) (taken from Van Valin, 2005: 201) is an
example of a core cosubordination construction (see also Bohnemeyer and
Van Valin 2017: 155f ). On the surface, the deontic modal operator -meli
(‘should, ought to’) is embedded in the second core, but it takes scope over
the entire complex core.

CLAUSE
CORE[MOD deont]
CORE[MOD deont] CORE[MOD deont]
NUC LM NUC OP[CO [MOD deont]] PRO
V ip V meli yiz
gid gör
Figure 20.33 Derived tree for (18)
(18) [[Gid-ip]CORE [gör-meli-yiz]CORE]CORE.

go-lm see-mod-1pl
‘We ought to go and see.’
Let us assume that the first core is added to the second core by sister
adjunction, similar to the analysis of the Lakhota construction in
Figure 20.27. The modal operator in (18) adjoins to the second embedded
core node and carries a feature indicating that it is a core operator. The
result should be the derived structure in Figure 20.33 with the mod feature
shared between all the three CORE nodes involved (cf. Van Valin, 2005:
204).16 In a cosubordination, an operator embedded in one part of the
complex structure generally takes scope over the larger category. Accord-
ingly, in all elementary trees for cosubordination configurations, the rele-
vant features (here mod) are shared between the lower and the higher
category in question (here the two CORE nodes). This is taken to be a general
property of cosubordination structures. Corresponding to this, we assume
that when mapping our derived structure to the standard RRG structure, the
operator targets the highest corresponding node, as long as there is no
higher operator level and no substitution node in between. In the case of
Figure 20.33, this is the core of the entire sentence.
Figure 20.34 gives the derivation of the tree in Figure 20.33, attaching the
tree for gid ip by sister adjunction. Note that this implies that there is a
special tree for cores such as gör yiz in this example, providing two CORE
nodes and requiring an adjunction at the higher one in order to be com-
pleted to a cosubordination structure. This requirement is again expressed
by an edge feature cosub. We can adjoin several cores such as gid ip but we
have to adjoin at least one in order to switch the feature from to þ.
Besides the edge feature, we also assume (as before, see Figure 20.27) a
boolean node feature cosub that expresses whether a node roots a cosubor-
dination structure or not (see also the CORE nodes in Figure 20.34). This
enforces an adjunction of the gid tree at the cosubordination CORE node and
not at the lower CORE node.

CLAUSE
COSUB + COSUB +
CORE ∗ COSUB + CORE
MOD 1 MOD 2
COSUB − COSUB −
COSUB + CORE COSUB − CORE
MOD 1 MOD 2
NUC LM NUC PRO
V ip V yiz
gid gör COSUB − ∗

CORE
MOD deont
OP[CO [MOD deont]]
meli
Figure 20.34 Derivation for (18)
A CORE cosubordination example that involves wrapping substitution

instead of sister adjunction is the English example in (19a) (cf. Van Valin
2005: 203), where we have a core consisting of three embedded core constitu-
ents where the first contains the modal operator must. This operator takes
scope over the entire large core. By contrast, in (19b) we have a structure
consisting of two cores which constitute a clause. That is, we have a core
coordination and not a core cosubordination. In this case, the modal embed-
ded in the first core scopes only over this one and not over both cores.
(19) a. [[Kim mustMOD go]CORE [to try]CORE [to wash the car]CORE]CORE
b. [[Kim mustMOD ask Pat]CORE [to wash the car]CORE]CLAUSE
Concerning the analysis of (19a), we propose that the different cores are
combined with each other via wrapping substitution and not via sister
adjunction. The difference, compared to (18) is that the verbs go and try
both syntactically select for an infinitival complement clause (i.e. for a
CORE argument), which should be expressed via a corresponding substi-
tution node. The derivation is shown in Figure 20.35. The node features
contributed by CORE operators (here [mod deont]) are shared between the
different CORE nodes that are part of the cosubordination construction.
An edge feature for enforcing the adjunction of further CORE sisters in a
cosubordination is not necessary since the substitution nodes guarantee
that CORE arguments have to be added.
The shared operator scope in (19a) is a standard criterion for distinguish-
ing cosubordinate from coordinate constructions. Another diagnostic is the

COSUB +
CORE
MOD 1
COSUB − COSUB −
CORE CORE
MOD 1 MOD 1
CLAUSE
RP NUC
COSUB +
CORE
MOD 2 COSUB +
Kim V CORE
MOD 3
go COSUB − COSUB −
CORE CORE
MOD 2 MOD 2 COSUB −
COSUB − ∗ CORE
CORE MOD 3
MOD deont
to try
OP[CO [MOD deont]] to wash the car
must
Figure 20.35 Derivation for (19a)
independent accessibility of the embedded cores by time-positional adver-

bials, which are analysed as core-level modifiers (cf. Bohnemeyer and Van
Valin 2017). While (19b) does allow independent time-positional modifica-
tion, as in (20a), this is not an option for (19a): both (20b) and (20c) are
excluded.
(20) a. Kim must ask Pat now to wash the car tomorrow
b. #Kim must go now to try to wash the car tomorrow
c. #Kim must go to try now to wash the car tomorrow
As to operators in subordinated CORE or CLAUSE arguments, since these

arguments are added by substitution, their root nodes block edge feature
percolation and we can have different operators within the argument and
outside. In (21) for instance (adapted from Van Valin 2005: 200), the tense
operator will is part of the argument CLAUSE added by substitution, and it
does not percolate upwards to the matrix clause.
(21) Kim told Pat that she will arrive late.
The substitution of the clausal argument into the tree anchored by told is
shown in Figure 20.36. The operator will in the embedded clause contributes
tense at the clausal layer. The corresponding feature [cl [tense fut]] is shared
along the clausal skeleton of the embedded sentence, via the ½op 2 features.
But it is not transported into the embedding clause, which can have a
different tense feature. The edge feature percolation also stops at the substi-
tution node, which means that the tense in the embedded clause cannot
satisfy a tense requirement in the embedding clause.

CLAUSE[OP [CL [TENSE past]] CLAUSE[OP 2 ]
CORE[OP [CL [TENSE past]] CLAUSE LM CORE[OP 2 ]
RP NUC[OP [CL [TENSE past]] RP that RP NUC[OP 2 ] MPperi
Kim told Pat she arrive late

CORE[OP 1 ]∗
⎡ ⎤
TNS +
⎣ CL_L + ⎦ OP 1 [CL [TENSE fut]] TNS −
OPS
CO_L +
will
Figure 20.36 Derivation for (21)
20.7 Modification and Periphery
Adjuncts in RRG are part of the periphery, which can be seen as a structure
similar to the operator projection since each adjunct targets a specific layer
and, starting from the nucleus and moving outside, the nuclear adjuncts
have to precede the core adjuncts, which in turn have to precede the clausal
adjuncts. Kallmeyer and Osswald (2017) briefly mention that this can be
modelled in a way similar to the treatment of operators explained above. In
the following, we will illustrate with an example how to capture the
ordering constraints for periphery elements via edge features in a way
analogous to the corresponding word-order constraints for operators. We
will restrict ourselves to adverbs but other modifiers would be treated in a
similar way.
According to Van Valin (2005: 41), adverbs may ‘modify all three layers
of the clause; aspectual adverbs like completely and continuously modify the
nucleus, pace adverbs like quickly and manner adverbs like carefully
modify the core, and epistemic adverbs like probably and evidential
adverbs like evidently modify the clause’. Depending on their position,
manner adverbs can actually be clausal or core modifiers, see (22), from
Van Valin (2005).
(22) a. Ruth cleverly hid the cash. (ambiguous)

b. Ruth hid the cash cleverly. (core modifier)
c. Cleverly, Ruth hid the cash. (clause modifier)
The core modifier reading signifies that Ruth hid the cash in a clever way
while the clausal modification has the meaning that it was clever of Ruth to
hide the cash. (22a) is ambiguous between the two readings while (22b)

seems to have only the ‘in a clever way’ reading and (22c) only the wide-
scope reading (‘it was clever to hide the cash’).
As to linear precedence, the main difference between modifiers and oper-
ators is that modifiers can often be placed to the left or to the right of the
nucleus (see (22)). Moreover, modifiers differ from operators regarding con-
straints on adjunction: they are rarely obligatory and there can be multiple
modifiers targeting the same layer, as in (23). If multiple modification is not
possible, this is due to semantic or pragmatic ill-formedness (e.g. conflicting
aspectual information).
(23) a. Ruth carefully hid the cash slowly. (several core modifiers)
b. Evidently, Ruth possibly hid the cash. (several clause modifiers)
In order to model the word-order constraints for periphery modifiers,

while going through the sentence in a direction away from the nuclear
predicate, we have to keep track of the modifier layer that we have already
reached. As an example, consider the possible placements of the adverbs
evidently (evidential, clausal periphery), slowly (pace adverb, core periphery)
and completely (aspectual adverb, nuclear periphery) within the sentence in
(24), an example taken from Van Valin (2005: 20).
(24) Leslie has been immersing herself in the new language.
The corresponding syntactic tree (without periphery elements) is given in

Figure 20.37. Note that here we choose a version where the aspectual
operator been, which scopes at NUC, attaches to CORE. This is what a parser
would choose in order to avoid crossing branches while being able to adjoin
an adverbial at CORE that is placed between been and immersing. All three
adverbs can be placed either to the left or to the right of the nucleus, and
combinations of this are also possible (cf. Van Valin 2005: 20). The position of
the adverbs has to be such that if more than one is to the left of the verbal
nucleus, within this group, clausal adverbs have to precede core adverbs,
which in turn have to precede nuclear adverbs, while for the group of
adverbs placed to the right of the verb, the opposite order has to be
respected (first nuclear, then core and then clausal adverbs).
CLAUSE
CORE
RP OP OP NUC RP PP
Leslie has been V herself in the new language
immersing
Figure 20.37 Syntactic tree for (24)

CLAUSE
CORE
RP PERIS 1 OP PERIS 1 PERIS 2 OP PERIS 2 NUC herself in …
Leslie has been V
immersing
CORE∗ CORE∗
PERIS [CL_L +] ADV PERIS [CO_L +] ADV PERIS [CL_L −]
evidently slowly
Figure 20.38 Derivation for (25a)
In order to control the order of modifiers with respect to the layered

structure, we assume an edge feature peris (periphery structure) similar
to the feature ops (operator structure) introduced in Section 20.5. The way
this feature controls modifier order is exactly as in the operator case. The
operator feature ops has of course to be percolated along the edge features
of the periphery trees and, similarly, operator trees have to percolate the
peris feature. Figure 20.38 illustrates how the example in (25a) is derived
using the peris edge feature for keeping track of the periphery structure
level reached.
(25) a. Leslie has evidently been slowly immersing herself in the new language.
b. *Leslie has slowly been evidently immersing herself in the new language.
Sentence (25b), by contrast, would be ruled out since the requirement to

have peris [cl_l ] when adjoining slowly would not be satisfied.
In addition to the edge features, which constrain the order of the per-
iphery elements, the ADV node itself can have a feature that indicates at
which layer it takes scope. For instance [scope cl] for evidently and [scope co]
for slowly.
20.8 Linking Syntax and Semantics
The main focus of the present chapter is on formalizing the syntactic side of
RRG. This section shows briefly how the described tree rewriting approach
can be extended to combining syntactic and semantic composition along
the lines of Kallmeyer and Osswald (2013) and Kallmeyer et al. (2016). In this
approach, the linking between syntactic and semantic components is

CLAUSE
CORE[IND e]
RP[IND x] NUC[IND e] RP[IND y]
Kim V[IND e] the vase
smashed
e = [do′(x, Ø)] CAUSE [BECOME smashed′(y)], x = Kim, y = vase
⎡ ⎤
causation
⎢ ⎡ ⎤ ⎥
⎢ activity ⎥
⎢ ⎣ ⎦ ⎥
⎢ person ⎥
⎢ x ⎥
⎢ ‘Kim’ ⎥
e⎢
⎢
⎡ ⎤⎥
⎥
⎢ change_of_state ⎥
⎢ ⎢ ⎥⎥
⎢ ⎢ y vase ⎥⎥
⎢ ⎢ ⎥⎥
⎣ ⎣ smashed_state ⎦⎦
y
Figure 20.39 Linking syntax and semantics: derived tree
encoded by an interface feature ind(ex), which can occur at nodes in the

syntactic tree. The value of this feature is an identifier, or label, which refers
to a specific component of the semantic representation.
Consider the simple example in (26) and its syntactic representation in the
upper part of Figure 20.39.
(26) Kim smashed the vase.
The figure illustrates how the syntactic constituents, represented by nodes

in the tree, are linked to components of the associated semantic representa-
tion. In fact, the figure shows two alternative semantic representations: a
logical structure of the type common in RRG (cf., e.g., Van Valin 2005: 151),
and a semantic frame, represented as an attribute value matrix, that captures
basically the same information about the event and its participants using
types, attributes and values. Possible ways of translating RRG’s logical struc-
tures into semantic frames are discussed in Osswald and Van Valin (2014)
and Osswald (2021). An advantage of using frames is that constraints on
semantic representations can be formalized in terms of types and attribute-
value constraints, and that semantic composition comes down to
frame unification.
In accordance with the general approach introduced in Section 20.3,
the syntactic tree in Figure 20.39 is derived by, first, combining elemen-
tary tree templates with lexical anchors and, second, combining the
resulting lexicalized elementary trees via substitution, sister adjunction

CLAUSE
CORE[IND e]
⎡ ⎤
RP[IND x] NUC[IND e] RP[IND y] causation
⎢ ⎥
⎢ activity ⎥
⎢ ⎥
V[IND e] ⎢ x ⎥
⎢ ⎡ ⎤⎥
e⎢
⎢ change_of_state ⎥
⎥
⎢ ⎢ ⎥⎥
⎢ ⎢ y ⎥⎥
V[IND e ] ⎢ ⎢ ⎥⎥
y
smashed
Figure 20.40 Anchoring the default transitive template with ‘smashed’
CLAUSE ⎡ ⎤
causation
⎢ ⎥
CORE[IND e] ⎢ activity ⎥
⎢ ⎥
⎢ x ⎥
⎢ ⎡ ⎤⎥
⎢
e⎢ change_of_state ⎥
⎥
RP[IND x] NUC[IND e] RP[IND y] ⎢ ⎢ ⎥⎥
⎢ ⎢ y ⎥⎥
⎢ ⎢ ⎥⎥
V[IND e]
y
smashed
RP[IND x ] RP[IND y ]
Kim the vase

person
x y vase
‘Kim’
Figure 20.41 Syntactic composition for (26) after argument linking
or wrapping substitution. For the given example, the grammar would

contain a transitive template (an elementary tree) that gets anchored by
the lexical item smashed (cf. Figure 20.40). The anchoring step (i.e. the
insertion of the smashed tree into the V◊ node) induces a unification of
the feature structures on the two V nodes. The linking between semantic
participants and syntactic argument slots can then be computed via a
constraint-based formulation of the linking algorithm along the lines of
Kallmeyer et al. (2016). The result of the lexical anchoring and the
argument linking is the elementary tree plus the semantic frame at the
top of Figure 20.41. The trees for Kim and the vase and their respective
frames can now be added by substitution, which means that the frames
labelled x′ and x unify and likewise the frames labelled y′ and y. The

result is the interlinked pair consisting of a syntactic tree and a semantic

frame shown in Figure 20.39.
20.9 Conclusion
The formalization of RRG syntax presented in this chapter puts emphasis on

drawing a clear line between declarative and procedural elements and on
the proper modelling of syntactic compositionality. Leaning on concepts
from Tree Adjoining Grammars, we formalized the syntactic dimension of
RRG as a tree rewriting grammar consisting of elementary tree templates
that are anchored by lexical items and that are combined by three modes of
composition: (simple) substitution, (sister) adjunction, and wrapping (sub-
stitution). Moreover, elementary tree templates are specified by classes of
tree constraints (in the so-called meta-grammar), which allows us to make
explicit how the different templates are structurally related to each other.
Employing wrapping substitution turns out to be vital for the appropriate
modelling of syntactic composition in RRG. Besides allowing the compos-
itional derivation of extraction from complement constructions from elem-
entary argument construction templates (see Section 20.3.3), wrapping
substitution is also well suited to deriving coordination and cosubordina-
tion chains, including those that can arise in embedded control and matrix-
coding constructions (see Section 20.6.1).
The proposed formalization does not introduce a separate tree structure
for representing the operator projection but treats operators as part of the
constituent tree. In order to avoid crossing branches, operators need not be
directly attached to the constituent nodes that define their scope-taking
behaviour. Rather, this information is encoded in node features associated
with the operator tree. An elaborate system of features is also responsible for
enforcing the correct surface order of the operators, which reflects their
scopal domain (see Section 20.5.2).
While the formalization of RRG’s semantic structures and its linking
system is beyond the scope of the present chapter, we sketched in Section
20.8 how semantic representations can be compositionally integrated with
the formal syntactic framework introduced here. We proposed the use of
semantic frames for this purpose, but the described interface between
syntax and semantics is open to other formal semantic approaches, includ-
ing a formalized version of RRG’s logical structures.
A complete formalization of RRG as a theory of grammar along the lines of
the present chapter requires additional steps, of course. In particular, RRG’s
linking algorithm should be fully spelled out as a system of constraints that
make reference to the chosen formal syntactic and semantic representa-
tions. Another important and non-trivial task is the formal representation of
information structure and its integration with the syntactic and semantic
representations.

References
Bladier, Tatiana, Andreas van Cranenburgh, Kilian Evang, Laura Kallmeyer,

Robin Möllemann and Rainer Osswald. 2018. RRGbank: A Role and Refer-
ence Grammar corpus of syntactic structures extracted from the Penn
Treebank. In Proceedings of the 17th International Workshop on Treebanks and
Linguistic Theory, 5–16. Linköping Electronic Conference Proceedings.
Bladier, Tatiana, Laura Kallmeyer, Rainer Osswald and Jakub Waszczuk.
2020. Automatic extraction of tree-wrapping grammars for multiple lan-
guages. In Proceedings of the 19th Workshop on Treebanks and Linguistic Theories,
55–61. Association for Computational Linguistics.
Bohnemeyer, Jürgen and Robert D. Van Valin, Jr. 2017. The macro-event
property and the layered structure of the clause. Studies in Language
41(1): 142–197.
Boyd, Adriane. 2007. Discontinuity revisited: An improved conversion to
context-free representations. In The Linguistic Annotation Workshop at ACL
2007, 41–44. Prague, Czech Republic.
Crabbé, Benoit and Denys Duchier. 2005. Metagrammar redux. In Henning
Christiansen, Peter Rossen Skadhauge and Jørgen Villadsen (eds.), Con-
straint Solving and Language Processing, Lecture Notes in Computer Science
3438, 32–47. Berlin: Springer.
Crabbé, Benoit, Denys Duchier, Claire Gardent, Joseph Le Roux and Yannick
Parmentier. 2013. XMG: eXtensible MetaGrammar. Computational Linguis-
tics 39(3): 1–66.
Croft, William, Jóhanna Barðdal, Willem Hollmann, Violeta Sotirova and
Chiaki Taoka. 2010. Revising Talmy’s typological classification of complex
events. In Hans C. Boas (ed.), Contrastive Studies in Construction Grammar,
Frank, Robert. 2002. Phrase Structure Composition and Syntactic Dependencies.
Cambridge, MA: MIT Press.
Ginzburg, Jonathan and Ivan A. Sag. 2001. Interrogative Investigations: The Form,
Meaning and Use of English Interrogatives. Stanford, CA: CSLI Publications.
Johnson, Mark. 1987. A new approach to clause structure in Role and
Reference Grammar. In Davis Working Papers in Linguistics 2, 55–59. Davis,
CA: University of California.
Joshi, Aravind K. 1985. Tree adjoining grammars: How much context-
sensitivity is required to provide reasonable structural descriptions? In
D. Dowty, L. Karttunen and A. Zwicky (eds.), Natural Language Parsing,
Joshi, Aravind K. and Yves Schabes. 1997. Tree-adjoining grammars. In Grze-
gorz Rozenberg and Arto Salomaa (eds.), Handbook of Formal Languages,
Vol. 3: Beyond Words, 69–123. Berlin: Springer.
Joshi, Aravind K., Laura Kallmeyer and Maribel Romero. 2008. Flexible compos-
ition in LTAG: Quantifier scope and inverse linking. In Harry Bunt and
Reinhard Muskens (eds.), Computing Meaning, Vol. 3, 233–256. Berlin: Springer.

Kallmeyer, Laura. 1999. Tree Description Grammars and Underspecified Representa-

tions. PhD thesis, Universität Tübingen. Technical Report IRCS-99-08 at the
Institute for Research in Cognitive Science, Philadelphia.
Kallmeyer, Laura. 2016. On the mild context-sensitivity of k-tree wrapping
grammar. In Annie Foret, Glyn Morrill, Reinhard Muskens, Rainer Oss-
wald and Sylvain Pogodalla (eds.), Formal Grammar: 20th and 21st Inter-
national Conferences, FG 2015, Barcelona, Spain, August 2015 (Revised Selected
Papers. FG 2016, Bozen, Italy, August 2016, Proceedings, number 9804 in
Lecture Notes in Computer Science), 77–93, Berlin: Springer.
Kallmeyer, Laura and Rainer Osswald. 2013. Syntax-driven semantic frame
composition in lexicalized Tree Adjoining Grammars. Journal of Language
Modelling 1(2): 267–330.
Kallmeyer, Laura and Rainer Osswald. 2017. Combining predicate-
argument structure and operator projection: Clause structure in Role
and Reference Grammar. In Marco Kuhlmann and Tatjana Scheffler
(eds.), Proceedings of the 13th International Workshop on Tree Adjoining Gram-
mars and Related Formalisms (TAGþ13), 61–70. Association for Computa-
tional Linguistics.
Kallmeyer, Laura, Rainer Osswald and Robert D. Van Valin, Jr. 2013. Tree
wrapping for Role and Reference Grammar. In Glyn Morrill and Mark-Jan
Nederhof (eds.), Formal Grammar (FG 2012/2013), Lecture Notes in Computer
Science 8036, 175–190. Berlin: Springer.
Kallmeyer, Laura, Timm Lichte, Rainer Osswald and Simon Petitjean. 2016.
Argument linking in LTAG: A constraint-based implementation with XMG.
In David Chiang and Alexander Koller (eds.), Proceedings of the 12th Inter-
national Workshop on Tree Adjoining Grammars and Related Formalisms
(TAGþ12), 48–57. Association for Computational Linguistics.
Kanazawa, Makoto. 2016. Multidimensional trees and a Chomsky-
Schützenberger-Weir representation theorem for simple context-free
tree grammars. Journal of Logic and Computation 26(5): 1469–1516.
Kroch, Anthony. 1987. Unbounded dependencies and subjacency in a Tree
Adjoining Grammar. In Alexis Manaster-Ramer (ed.), The Mathematics of
Language, 134–172. Philadelphia: John Benjamins.
Nolan, Brian. 2004. First steps toward a computational RRG. In Proceedings of
the International Role and Reference Grammar Conference 2004, 196–223. Insti-
tute of Technology Blanchardstown, Dublin.
Osswald, Rainer. 2021. Activities, accomplishments and causation. In Robert
D. Van Valin, Jr. (ed.), Challenges in the Analysis of the Syntax–Semantics–
Pragmatics Interface: A Role and Reference Grammar Perspective, 3–30. Newcastle
upon Tyne: Cambridge Scholars Publishing.
Osswald, Rainer and Laura Kallmeyer. 2018. Towards a formalization of
Role and Reference Grammar. In Rolf Kailuweit, Eva Staudinger and
Lisann Künkel (eds.), Applying and Expanding Role and Reference Grammar
(NIHIN Studies), 355–378. Freiburg: Albert-Ludwigs-Universität,
Universitätsbibliothek.

Osswald, Rainer and Robert D. Van Valin, Jr. 2014. FrameNet, frame struc-
ture, and the syntax–semantics interface. In Thomas Gamerschlag, Doris
Gerland, Rainer Osswald and Wiebke Petersen (eds.), Frames and Concept
Types, 125–156. Dordrecht: Springer.
Osswald, Rainer and Robert D. Van Valin, Jr. 2022. The description of transi-
tive directed motion in Lakhota (Siouan). In Laure Sarda and Benjamin
Fagard (eds.), Neglected Aspects of Motion Events Description: Deixis, Asymmetries,
Constructions, 209–233. Amsterdam: John Benjamins.
Pollard, Carl J. and Ivan A. Sag. 1994. Head-Driven Phrase Structure Grammar.
Rambow, Owen. 1994. Formal and Computational Aspects of Natural Language
Syntax. PhD thesis, University of Pennsylvania.
Rambow, Owen, K. Vijay-Shanker and David Weir. 1995. D-tree grammars. In
Proceedings of the 33rd Annual Meeting of the Association for Computational
Linguistics, 151–158. Association for Computational Linguistics.
Rambow, Owen, K. Vijay-Shanker and David Weir. 2001. D-tree substitution
grammars. Computational Linguistics 27(1): 87–121.
Sag, Ivan. 2012. Sign-Based Construction Grammar: An informal synopsis. In
Hans Boas and Ivan Sag (eds.), Sign-Based Construction Grammar, 61–188.
Sag, Ivan A. and Thomas Wasow. 1999. Syntactic Theory: A Formal Introduction.
Ullrich, Jan (ed.) 2011. New Lakota Dictionary (2nd ed.). Bloomington, IN:
Lakota Language Consortium.
Ullrich, Jan. 2018. Modification, Secondary Predication and Multi-Verb Construc-
tions in Lakota. PhD thesis, Heinrich Heine University Düsseldorf.
Notes
1 Round brackets indicate optionality, curly brackets mean that the

category in question may not be available in every language. DP stands
for ‘detached position’ and ECS for ‘extra-core slot’ (covering pre- and post-
core slots).
2 The given analysis is slightly simplified in that the operator did contrib-
utes also to other layers than CLAUSE; see Section 20.5 for a more detailed
exposition, which also covers operators that are realized as bound
morphemes.
3 The possible occurrence of ‘crossing branches’ due to mismatches
between the surface position of an operator or a periphery element and
its scope will be discussed in Sections 20.5 and 20.7.

4 Sister adjunction has been introduced as a composition operation on so-

called d-trees by Rambow et al. (1995). Notice that sister adjunction differs
from adjunction in TAG. In fact, the latter operation is more akin to the
wrapping operation described in Section 20.3.3.
5 We will see later how edge features can be used to constrain the positions of
operators or periphery elements among the daughters of their target nodes.
6 This analysis is inspired by TAG, where the slot for the wh-phrase and its
predicate originate from the same elementary tree and the elementary
tree of the embedding verb adjoins in between; see Kroch (1987).
7 Wrapping substitution is related to the concept of flexible composition
proposed in Joshi et al. (2008), which allows one to interpret TAG adjunc-
tion as a wrapping operation.
8 The idea of d-edges or dominance links has been used in various exten-
sions of TAG, for instance in V-TAG (Rambow 1994), Tree Description
Grammars (Kallmeyer 1999), and D-Tree-Substitution Grammar (Rambow
et al. 2001).
9 Bladier et al. (2020) propose a slight extension of wrapping substitution
where in cases of an empty tree γ, the upper node targeted by the
operation need not be a root node.
10 This means that for a given grammar, one can implement a parsing
algorithm such that for a sentence of length n, the program computes
in at most cnm steps whether the sentence belongs to the language, c and
m being fixed constants.
11 This transformation of splitting a node with a discontinuous span into
several nodes for the different components was also proposed by Boyd
(2007), though not with a unique identifier for the split node.
12 A similar constraint on elementary trees is often proposed in TAG-based
approaches to linguistic analysis; cf., e.g., Frank (2002).
13 Since in passive constructions, the auxiliary be is structurally required
for nucleus formation, it has a representation in the constituent projec-
tion as an AUX node; cf. Van Valin (2005: 13, note 3).
14 A computational implementation is provided by the meta-grammar
compiler XMG (eXtensible MetaGrammar, Crabbé et al. 2013), which
allows specifications of RRG trees including d-edges (xmg.phil.hhu.de).
The resulting grammars can then be processed using the TuLiPA parser
(github.com/spetitjean/TuLiPA-frames).
15 Such a representation is chosen in the RRGbank (rrgbank.phil.hhu.de)
and RRGparbank (rrgparbank.phil.hhu.de), both treebanks of syntactic
RRG structures, see Bladier et al. (2018).
16 We slightly simplify here because the mod feature would actually be
embedded under features op and co, for instance [op[co[mod deont]]].

21
Computational
Implementation and
Applications of Role and
Reference Grammar
Brian Nolan
21.1 Introduction
Functional linguists of all persuasions consider that theories of linguistic

structure should be directly relatable to testable theories of language
production and comprehension. Computational models assist in this pro-
cess, and today professional linguists can be found working in many areas
of knowledge and skills using strategies and techniques that did not exist a
generation ago. This chapter therefore examines the range of computa-
tional implementations and applications of Role and Reference Grammar
(RRG) and the characteristics of RRG that makes this functional model of
language amenable to implementation in software. We examine the elem-
ents of RRG treated in software: the architecture of the lexicon, the linking
system, the functional philosophy encapsulated within speech acts in a
‘language as communication’ perspective and the logical structures that
provide a meta-representation of an utterance. Important also are the
layered structure of the clause and noun phrase. Constructions in RRG
have also received a computational treatment. RRG has been used success-
fully to characterize languages as diverse as Irish, Persian, Maori, Arabic
and Japanese, along with many accusative and ergative languages, while
the languages that have undergone a computational treatment in RRG
have included English, Modern Standard Arabic, Spanish, Biblical Hebrew,
German, and others.
The chapter, then, is a testimonial to the potential of an RRG model of
language to be implemented in software, and to be useful in tasks such as
machine translation. It includes some discussion of what must be trans-
formed and augmented in an RRG account of a phenomenon for computa-
tional purposes, and why computer scientists have come to RRG in

786 BRIAN NOLAN
particular for their needs. The RRG model is rapidly evolving to new levels of
descriptive, explanatory and computational adequacy. By computational
adequacy we mean the degree to which a theoretical model can be formally
specified to enable computational modelling as a benchmark for that
theory. As will become clear, we find that the RRG model is most suitable
as a basis for computational implementation in software and can serve as a
rich testing ground for the linguistic theory. Computational models of RRG,
by intention linguistically motivated, deal with problems at the interfaces
between concept, semantics, lexicon, syntax and morphology, including
language processing and grammar. The challenge is to determine how best
to design and implement the model and its principled interfaces
in software.
What makes RRG suitable for implementation in software is that it is a
model of grammar that posits a direct mapping between the semantic
representation of a sentence and its syntactic representation (Van Valin
2005, 2008). RRG is a monostratal theory, positing only one level of syntac-
tic representation. The actual form of the sentence and its linking algo-
rithm can work in both directions from syntactic representation to
semantic representation and vice versa. In RRG, semantic decomposition
of predicates and their semantic argument structures are represented as
logical structures. The lexicon in RRG takes the position that lexical entries
for verbs should contain unique information only, with as much infor-
mation as possible derived from general lexical rules. RRG makes use of a
set of thematic roles organized into a hierarchy in which the highest-
ranking roles are Actor (for the most active participant) and Undergoer
(for most affected).
Amongst the design challenges is the requirement for a digital model
of the lexicon as a persistent store, either as a database, persistent XML
tree structure or serializable object. The architecture of the lexicon is
then an empirical problem for the software designer. It is clear that while
the RRG lexicon is traditionally considered with respect to verbal logical
structures, the software typically needs the lexicon to also include nouns,
adjectives, prepositions, and so on, in some meaningful way. To deal with
many of the problems found in modelling language within the model,
constructions also need to be stored as occurrence patterns. Additionally,
lexemes need to be meaningfully stored within the lexicon. As grammat-
ical morphemes are very much language-specific, a morpheme store is
also needed within the software. All the software and computing applica-
tions mentioned in the case studies make extensive use of the bidirec-
tional linking system from morphosyntax to semantics and semantics to
morphosyntax. The implementation may follow construction-oriented
processing according to some language-specific set of rules translated
into computer code.
The reasons why many computer scientists and computational linguists
use RRG are precisely because of the desire to deploy a functional linguistic

Computational Implementation and Applications 787
model that has a strong theory of the lexicon and a role for constructions.
Important also is the idea of projecting from the lexicon in a bidirectional
linking system, the layered structure of the word (LSW), the layered struc-
ture of the noun phrase (LSNP), and the layered structure of the clause with
its sub-model of nexus–juncture relations. The approach of RRG to argument
realization is highly principled in a typologically coherent way and this adds
considerable value to all language modelling efforts in software. Addition-
ally, the interfaces within RRG that reach from lexicon through semantics
via morphosyntax into speech act, information structure, pragmatics and
discourse are considered to be highly valuable.
The RRG linguistic model strives for high levels of descriptive, explana-
tory, predictive and typological adequacy (Butler 2009). Through the
deployment of the RRG model in computational applications it is shown
to meet computational adequacy. In many respects with regard to compu-
tational implementation, RRG can be considered as a ‘functional genera-
tive’ model with parse and generate ‘directions’ within the bidirectional
linking system. It does not, however, express itself as treating the
Chomsky-typed languages as formal constructs. It is more typologically
sensitive than that. While RRG can be modelled within procedurally
described or declarative approaches, it is always as a semantically motiv-
ated and lexically projected model of language. It may make use of
attribute-value matrices (AVMs) and constructions, depending on the
object of study within the computational model.
The chapter has the following organIsation. In Section 21.2, we charac-
terize the start of computational work on RRG, concentrating on the
syntactic side of the linking system. We look at the use of general-
purpose programming languages to create parsers for English, German,
Dyirbal and Biblical Hebrew. In Section 21.3, we give a description of an
RRG-based machine translation engine for Modern Standard Arabic that
parses a sentence in the native Arabic orthography and translates it into
grammatical English. The software employs an interlingua bridge archi-
tecture. Section 21.4 outlines current work on intelligent conversational
agents and humanoid avatars, where RRG is used as the language engine.
In Section 21.5, the FunGramKB project and related toolsets oriented
towards the RRG model are discussed. In Section 21.6, the contribution
of RRG to linguistic modelling in natural language processing (NLP)
applications is discussed.
21.2 The Computational Modelling of RRG
Computational work using RRG has been ongoing since 2004 in a variety of
ways (Nolan 2004). Initial concerns with computational treatments of RRG
were with exploring the possibilities of expressing the theory in a compu-
tationally tractable way as a feature-based unification grammar. This

788 BRIAN NOLAN
involved creating a hierarchy of types, each of which was structured as an

attribute-value matrix with a set of features and associated values. A set of
rules was created and lexicon entries were structured to unify with the
rules, somewhat along the lines of a phrase structure grammar. Early
modelling was achieved using the constraint-based computer program-
ming language Prolog, treating elements of RRG as a definite clause gram-
mar. In this approach, the lexical rules and the lexical entries shared a
common formalism. However, it proved not to be feasible to formalize the
full RRG model using Prolog, as the RRG theory is not a context-free
grammar. It also proved not to be possible to use Prolog to implement
the operator projection, the linking system, the rich lexicon, and the full
layered structure of the clause as understood in RRG. Later, additional
modelling was undertaken using a software toolset, based on Common
Lisp, called the Linguistic Knowledge Builder (LKB). The LKB is an open-
source grammar engineering environment for creating grammars and
lexicons of natural languages, for typed feature structure grammars
(Copestake 2002). LKB is typically used for grammars with typed feature
structures, such as Head-Driven Phrase Structure Grammar (HPSG), with
unification. The type system constrains the allowable structures and pro-
vides a way of capturing linguistic generalizations. At that time, the Prolog
language and the LKB system were seen to offer several advantages in
potentially making computational linguistics and natural language
processing accessible to linguists with a very limited background in com-
puter science. While the LKB system was successfully used to build and
model small RRG grammars of English, these remained at the level of
small-scale experiments. Overall, the LKB proved to be unwieldy for hand-
ling the core theoretical elements of RRG, such as the design of the lexicon,
and the linking system. Following the limitations found in LKB and the
complexity of large-scale Prolog programs, development work on RRG-
based computational linguistics progressed to more modern general-
purpose programming languages. These languages also offered the power-
ful advantage of enabling the creation of graphical user interfaces (GUIs) to
generate dynamic visualizations of RRG trees of the layered structure of
the clause more easily. They also offered the opportunity to create more
robust applications that implement the RRG linguistic model along with
more sophisticated data structures and XML.
Guest (2008: 435–453) reports on RRG-based software developed for parsing
a selection of sentences from Dyirbal (including sentences with discontinu-
ous constituents and free word order), English (fixed word order), and a
small sample of Dutch. This parser exploits two features of RRG: first, that
syntactic structure is represented in terms of templates instead of rules, and
second, that grammatical categories (operators) are represented differently
from predicates, arguments and adjuncts. Guest extends a chart parser
strategy to enable varying degrees of word-order flexibility and parsing via
templates instead of rules. The choice of templates rather than rules was

made because templates enable more information to be captured, which in

turn makes it easier to extract the meaning from the sentence.
From Guest’s (2008: 435) perspective, RRG is a suitable theory for a compu-
tational treatment for several reasons.
1. RRG posits constituent, operator, and focus projections, such that the
various aspects of a sentence can be dealt with separately. This enables
words that modify other words to be removed from the constituent
projection and placed in an operator projection, which shows how these
words modify the meanings of the words in the constituent projection.
Therefore, only the main constituents of a sentence have to be parsed,
simplifying the parsing process.
2. Secondly, RRG has a strong bidirectional linking system from syntax to
semantics. Grammatical constructs are designed to be both cross-
linguistically valid and to make the meaning relatively easy to extract.
Conversely, she finds there are some elements of RRG which make it harder
to implement. Guest (2008: 436) notes these to include:
1. It is much more difficult to parse with templates than with rules, and
RRG templates are particularly hard because, in a parse tree, lines are
allowed to cross. The parse trees are not made up of parents and
children, and nodes can have modifiers (such as PERIPHERY) attached
to them.
2. In RRG, word-order constraints are implicit in the templates. The
theory supports modelling of examples from many languages which
include fixed and free word order, and mixed varieties in between.
Parsing algorithms tend to handle either fixed-word-order or free-word-
order languages. Most parsing algorithms, such as those based on
HPSG and context-free grammar, are based on rules. Although tem-
plates can easily be reduced to rules, information is lost in going
this route.
As we mentioned, Guest (2008: 436) describes an approach based on a
chart parser that has been modified to handle both templates and varying
degrees of word-order flexibility. This parsing algorithm is characterized by
Guest as consisting of several stages:
1. Use Toolbox1 to tag the sentences.
Toolbox, a semi-interactive tagging program, was chosen by Guest
because: (i) A user can define their own tags and (ii) it is easy to
ensure tags are correct. Toolbox will ask the user if there is more
than one possibility for a word.
2. Strip away the operators.
This removes all words that modify other words, based on a correct
tagging of head and modifying words. The modifying tags that
belong to head words are specified in a file, and the end result is a
simplified sentence to be input to the parser.

790 BRIAN NOLAN
3. Parse the simplified sentence using templates.

Guest achieves this by collapsing the templates to rules, parsing using a
chart parser and then rebuilding the trees at the end. The chart parser
has been modified to handle varying degrees of word-order flexibility.
4. Draw the resulting parse tree.
The parser has been tested by Guest on example sentences from Dyirbal,
English and Dutch. This proof-of-concept parsing method for RRG, using
templates, can handle both fixed word order and variable word order with-
out any difficulty.
Diedrichsen (2014: 105–142) describes an implementation of a sentence
parser for German that implements the linking system of RRG from syntax
to semantics (see also Diedrichsen 2008; Van Valin and Diedrichsen 2006),
and an implementation of the RRG lexicon in a data structure. This
RRG parser for German employs the concept of a construction as a
Table 21.1 The German bracket structure construction (based on Diedrichsen

2014: 109)
CONSTRUCTION: German bracket structure construction

SIGNATURE: VFIN [ARG1. . .ARG2. . .ADJ1. . .PARTICLE] PART.OF.VERBAL.EXPRESSION
Space between brackets may be empty (if clause is intransitive)
CONSTRAINT: No constraints or restrictions.
WORKSPACE: Real-time processing according to the following construction-specific rules
SYNTAX:
Bracket Structure: Bracket is ALWAYS opened by the finite verb. The right bracket is filled according to
the following rules:
1. If VFIN ¼AUX
{PART.OF.VERBAL.EXPRESSION ¼ full verb (PSTP)};
2. If VFIN ¼full verb
{PART.OF.VERBAL.EXPRESSION ¼ separable prefix};
3. If VFIN ¼light verb
{PART.OF.VERBAL.EXPRESSION ¼ rest of collocation};
Order of arguments within brackets follows topic comment structure and principle of end
weight. The following rules apply:
1. General constraints: pronoun > other, RP > PP
2. Case-based argument ordering constraint: NOM > DAT > ACC (default)
3. If ACC ¼ pronoun, then ACC > DAT (default)
MORPHOLOGY:
AUX: May be any auxiliary or modal verb
VFIN may be any verb.
PHONOLOGY: Not specified.
SEMANTICS: [þtelic], where telicity may be only invoked for conversational purposes: adds intensity,
expression
PRAGMATICS:
Turn taking: Signal for dimension of Turn Constructional Unit (TCU);
right brace marks end of TCU
Illocutionary force: not specified
Focus structure inside braces: TOP>FOC

language-specific structural pattern to guide its operation over the ‘bracket

structure’ of the German clause (Table 21.1). The construction is considered
to be a grammatical object (Nolan 2013; Diedrichsen 2014: 109). Inside these
brackets, the main information of the sentence is placed in an order that
reflects the information structure. The software accepts an input sentence in
German and parses it into its respective tokens. Each token is checked to
determine its lexical category and the morphological marking of each token
is unpacked for its feature set. The parse captures the structural representa-
tion of the clause and grammatical features like tense in RRG notation. The
parser is able to recognize and parse eight construction types of the Stand-
ard German clause. Included are conventional sentences using active transi-
tive and intransitive structures with V2 and V1 patterns. The word order is
varied such that the direct object may appear in sentence-initial position.
The structure is parsed in terms of the layered structure of the clause, which
involves a nucleus, a core with arguments, a pre-core slot and a clause layer
that subsumes the other elements. The sentence structures are presented
within a GUI as a series of embedded frame structures according to the style
of attribute-value matrices (AVMs). Specifically, this style of representation
has the format of [feature: value] and value can contain nested AVMs such
as [feature: [feature: value]].
This parser is the first application able to parse German sentence construc-
tions according to RRG rules and provide an elaborate set of features with
each of the tokens from the constructions. German grammar is considered
to be notoriously difficult for computational applications, as it has (i) a
variety of forms in its lexicon set, due to the (ii) three genders and (iii) its
rich morphology, and it also has (iv) great freedom in word order, of which
the only regularity is that with declarative clauses, the finite verb has to
appear in the second position within the order of constituents. This parser
from Diedrichsen, which does not claim full coverage, is a valuable
approach to a full computational representation of German, and serves as
a robust proof of concept to demonstrate again that the RRG linguistic
model is amenable to computational implementation.
RRG-related computational work has been carried out on the parsing of
Biblical Hebrew (Winther-Nielsen 2008). This work involved the creation of a
web-based software application, called the Role-Lexical Module (RLM), to
build a lexicon of logical structures and semantic representations for RRG.
The goal of the RLM project has been to develop and test new methods for
lexical analysis within RRG in the analysis of Biblical Hebrew. Specifically,
this software is a computational tool for doing a semantic analysis of
Biblical Hebrew verbs in a text and organizing the results in an online
lexical database. The RLM tool uses an electronic database for Hebrew, and
a linguist can choose any clause from Genesis 1–3, with simple translation
glosses, and can then perform various computer-aided analyses. The RLM is
able to display the nodes of the RRG syntax trees representing the layered
structure of the clause. For computational work on the RLM tool, the RRG
theory offered the advantage of a bidirectional linking between syntactic

792 BRIAN NOLAN
structure and the representation of semantic structure in the lexicon

through mapping rules. The RRG theory supports a clear, semi-formal repre-
sentation of meaning and structure in all languages. The linguist can
classify verbs into Aktionsart classes, represent the meaning of clauses
following a selection of logical structure, and gradually build an RRG
lexicon for Biblical Hebrew. The tool stores the basic lexical meanings of
verb classes in the semantic metalanguage of RRG and it offers a lexical
representation of meaning in actual linguistic contexts through a WordNet
glossing. This tool is used for parsing, lexical representation, and verb class
decomposition. The RLM tool offers a WordNet lexical selector, a text display
of transliterated and glossed Hebrew, a syntactic parser for RRG, and a
logical structure decomposer.
Related to this project is the work of Gottschalk (2014), who developed a
computable account of RRG based on constructional schemata. Gottschalk’s
approach places emphasis on thematic relations within three-place predi-
cates rather than the semantic macroroles of actor and undergoer. In
Gottschalk’s view, the standard linking algorithm of RRG has been compu-
tationally enhanced for variable undergoer linking in English three-place
predicates. An innovation of Gottschalk’s computational work is the use of
conceptual graphs to implement logical structures in software. Gottschalk
implements conceptual graphs as an approach to the semantic component
of RRG. These have been applied within the Biblical Hebrew semantic parser
as part of the RLM, as a functional extension. The application of conceptual
graphs has the advantage that computational approaches to ancient lan-
guages such as Biblical Hebrew can be more easily developed (Gottschalk
2019). In this approach, a linking algorithm from syntax to semantics is
reduced to a set of lexical rules which match AVMs defining the layered
structure of the clause against an ontology of Biblical Hebrew.
21.3 Machine Translation: An RRG Interlingua
Machine translation (MT) is a subfield of computational linguistics that investi-

gates the use of computer software to translate text (or speech) from one
natural language to another. In a software application on a larger scale to
those already mentioned, the RRG linguistic theory has successfully been used
to underpin the language model in a lexically based proof-of-concept machine
translation system called UniArab (Nolan and Salem 2010, 2011) that supports
fundamental aspects of Arabic, including lexical category, agreement and
tenses. UniArab is based on the bidirectional linking algorithm of RRG (syntax
to semantics and vice versa). UniArab takes Modern Standard Arabic as
input in the native orthography, parses the sentence(s) into a logical meta-
representation based on the fully expanded RRG logical structures and gener-
ates perfectly grammatical English output with full agreement and morpho-
logical resolution. UniArab utilizes an XML-based implementation of elements

of RRG in software. In the analysis of Arabic within the software, the lexical
and grammatical properties of the Arabic words are extracted automatically.
From the parse, UniArab then creates a computer-based representation for the
logical structure of the Arabic sentence(s).
In this application, RRG has been used to motivate the computational
implementation of the lexicon as an XML data structure and to implement
in software the RRG bidirectional linking system to build the parse and
generate functions between the syntax–semantic interfaces. Through seven
input phases, including the morphological and syntactic unpacking, Uni-
Arab extracts the logical structure of an Arabic sentence. Using the XML-
based metadata representing the RRG logical structure, UniArab then accur-
ately generates an equivalent grammatical sentence in the target language
of English through four output phases. Nolan and Salem (2011) report on the
development of this machine translation system as a rule-based semantic-
ally oriented interlingua bridge framework for machine translation of
Arabic language processing using the RRG linguistic model. The UniArab
system is able to analyse intransitive, transitive and ditransitive Arabic
sentences in native Arabic (right to left) orthography and extract their
logical structure. Through a detailed study of the Arabic language, an
analyser was developed that can successfully process many of the unique
features and challenges present in Arabic. This logical structure is then used
in the generation phase, where the sentence is translated into another
language, in this case, English. The Arabic language is written from right
to left and it has complex, language-specific grammar rules and a relatively
free word order. These distinguishing features pose a major challenge in
processing Arabic text for linguistic analysis. The UniArab framework dem-
onstrates that RRG is a foundation for building multi-language machine
translation systems.
It is common to categorize the three different MT methodologies as direct,
transfer and interlingua. These levels of analysis are illustrated by what has
become known as the Vauquois triangle (Vauquois 1968) (see Figure 21.1). The
methodologies differ (i) in the linguistic depth of analysis of the source
language and (ii) in the degree to which they attempt to reach a language-
independent representation of linguistic meaning of the source and target
languages. An interlingua is designed to be a language-independent represen-
tation from which translations can be generated to different target languages.
According to Dorr (1992: 46), in order to adopt an interlingua approach to
machine translation, one must construct a language-independent represen-
tation that lends itself readily to the specification of a systematic mapping
that operates uniformly across all languages. Dorr describes such a repre-
sentation based on the lexical conceptual structure of Jackendoff (1990),
which abstracts away from syntax sufficiently to enable language-
independent encoding, while retaining enough structure to be sensitive to
the requirements for language translation. Machine translation using an
interlingua approach typically involves semantic analysis of the source

794 BRIAN NOLAN
Figure 21.1 The Vauquois triangle (based on Dorr et al. 2006: 384)
language (Dorr et al. 2006: 1). At the interlingua level, a single underlying
meta-representation represents the underlying semantics of the source lan-
guage SL1 text., which will be used to generate a translation into a target
language TL1. In principle, an interlingua representation of a sentence
contains sufficient information to allow generation in any language; the
more target languages there are, the more valuable an interlingua becomes.
When there are many source languages (SLn) and many target languages
(TLn) for translation, an interlingua bridge is deployed. This has an advantage
in facilitating the scaling up to multiple languages. To add a language to an
interlingua bridge, one needs to add, for that language, a lexicon, a parse
function into a meta-representation, and a generate function from the meta-
representation. To translate from one source into N target languages, one
needs (1 þ N) steps using an interlingua. However, to translate pairwise among
all the languages, one needs only 2N steps using an interlingua compared to
about N2 with transfer – this is a significant reduction in development, design
and processing cost. In addition, it is not necessary to consider the properties
of any other language during the analysis of the source language or gener-
ation of the target language; each analyser and generator can be built inde-
pendently by a monolingual development team. Each system developer only
needs to be familiar with their own language and the interlingua. An inter-
lingua MT system will include one lexicon for each language. An advantage of
the interlingua approach is that interlingua representations have the poten-
tial to be used by other multilingual NLP applications. More importantly,
using the logical structures of RRG as the interlingua meta-representation

provides distinct advantages, especially given the strong, justified claims of

RRG’s typological adequacy. Another advantage of using the logical structures
of RRG as the interlingua meta-representation is that the MT is immediately,
and by design, linguistically motivated.
UniArab uses the interlingua bridge approach (Figure 21.2) in its architec-
ture that allows for language extensibility within the software. The UniArab
MT model (Nolan and Salem 2011: 320) thus has two main stages: input-
parse-analysis and output-generate. This interlingua-based MT is done via
an intermediate semantic representation, based on RRG logical structures,
of the source language text. Each language therefore needs a bridge from
that language source to the interlingua logical structure [parse source to
LS] and a bridge from the interlingua logical structure back to the language
[generate target from LS] (Figure 21.3). This approach allows the soft-
ware to scale up should additional languages be added beyond the initial
source (SL1) and target (TL1) language pairs, and avoid a translation com-
plexity problem between languages. The implementation of the interlingua
bridge architecture solves the translation complexity problem as automatic
language translation is made from a source language into a kernel
Figure 21.2 MT–RRG-based interlingua bridge approach (from Nolan and Salem 2011: 315)
Figure 21.3 The RRG interlingua for UniArab (from Nolan and Salem 2011: 315)

796 BRIAN NOLAN
meta-representation (the input parse phase) and generates to a target

language from the meta representation (the generate phase).
Ambiguity problems for an interlingua in a multilingual system are still
likely if one of the languages involved has two or more potential forms for a
single given word in one of the other languages. However, a semantically
oriented approach to MT can potentially disambiguate more easily than
other strategies. For an interlingua to be completely language-neutral, it
must represent not the words of one or another of the languages but rather
language-independent lexical units. Any distinction that can be expressed
lexically in the languages of the system must be represented explicitly in the
interlingua representation. In this software, the RRG logical structures are
strategically used as the basis of the meta-representation in the interlingua
bridge with a lexicon encoded in XML. The UniArab system can generate the
target language output through classifying every Arabic word in the input
source text and creating a meta-representation of the sentence(s) input as a
text in a fully populated RRG-style logical structure, including the various
nominals and their associated features of [defþ, mascþ], and so on. Arabic
is considered to have six major ‘parts of speech’. These are verbs, nouns,
adjectives, proper nouns, demonstratives and adverbs. The major parts of
speech in the Arabic language have their own attributes, and the UniArab
software has a strong analytical component that can extract all these attri-
butes from the words in a sentence.
The UniArab system accepts Arabic as its source language (Figure 21.4).
The morphology parser and word tokenizer have a connection to the
Figure 21.4 The conceptual architecture of the UniArab system (from Nolan and Salem 2011: 321)

lexicon, which holds all attributes of a word. UniArab was developed with
the lexicon encoded in XML, and also stores all data in XML format. The
system can understand the part of speech of a word, agreement features,
number, gender and the word type. The syntactic parse unpacks the agree-
ment features between elements of the Arabic sentence into a semantic
representation (the logical structure) of the state of affairs of the sentence.
The structure of the UniArab system and its operational phases are
described next.
Phase 1. Input of Arabic language sentence: The input to the system consists
of one or more sentences in Arabic. UniArab can accept as input
one or multiple sentences in a single input.
Phase 2. Sentence Tokenizer: Tokenization is the process of demarcating
and classifying sections of a string of input characters. In this
phase, the system splits the text, which may consist of paragraphs,
into sentence tokens. The resulting tokens are then passed to the
word tokenizer phase.
Phase 3. Word Tokenizer: In this phase, sentences are split into word tokens.
The Arabic sentence is read from right to left.
Phase 4. Lexicon XML Data-source: The lexicon is constructed as a set of XML
documents for each component category of Arabic. The lexicon
contains the Arabic lexemes in this XML data source ordered by
each lexical category.
Phase 5. Morphology Parser: This works directly with both the lexicon and
tokenizer. A connection is made to the lexicon data-source of
Phase 4. To understand the morphology of each word, the software
tokenizes each sentence and determines the word relationships.
This phase of the system holds all attributes specific to each word
of the source sentence.
Phase 6. Syntactic Parser: Determines the structure of the Arabic sentence.
At this point, the types and attributes of all the words in the
sentence are known.
Phase 7. Syntactic linking (RRG): The software develops the link from syntax
to semantics out of the phrasal structure created in Phase 6 to
create a logical structure for later use in generation into the target
language, and also to act as the link in the opposite direction from
semantics to syntax.
Phase 8. Logical Structure: The creation of logical structure is the most
crucial phase. An accurate representation of the logical structure
of an Arabic sentence is the primary strength of UniArab.
Having described the various interlingua parse phases, we now characterize

in more detail the generate phase from the meta-representation (i.e. the
populated logical structure) to the target language (Nolan and Salem 2011:
323). The target language generation phases in the UniArab system follows a
syntactic realization model based strongly on RRG. Generation takes, as

798 BRIAN NOLAN
input, the universal logical structure of the input sentence(s), and produces,
as output, the grammatically correct morphosyntax of the target language.
In the UniArab system, Phases 9 to 12 are for generation of the target
language, which, in this case, is English.
Phase 9. Semantics to Syntax: Once the software has an input sentence and
has produced a structured syntactic representation of it, the
grammar can map this structure from a semantic representation.
In this phase, the system uses the linking algorithm of RRG to
determine Actor and Undergoer assignments, assign the core
arguments and assign the predicate in the nucleus. It determines
the grammatical ‘subject’ by analysing the agreement marking on
the verb and the various nominals. The system hinges on the use
of logical structures at the metalevel.
Phase 10. Syntax Generation: The generation phase from the interlingua
bridge meta-representation to the morphosyntax of a particular
target language will, of course, depend on the characteristics of
the target language. The generation phase implements the RRG
semantics-to-syntax linking system.
Phase 11. Generate English Morphology: The system generates English
morphology. There are some ‘tricky’ special cases, handled well
by the UniArab system, such as adding the copula verb ‘to be’ into
the English copula sentence, or changing the source language
verb’s tense to an appropriate and grammatically correct tense in
the target language, depending on the tense distinctions found
within the target language. Also, the word order in the target
language is considered and applied correctly.
Phase 12. English Sentence Generation: The process of generating an English
sentence can be as simple as keeping a list of rules, and these rules
can be extended through the life of the MT system. The system
applies some operations in English, such as vowel change in
the lexical item of English to denote sg vs. pl, for example man
vs. men. Sometimes this accompanies affixation: break/broke/broken
(¼brokeþen) to denote various tense and aspect distinctions.
This proof-of-concept software generates grammatically correct English. This

computational work has shown that RRG facilitates the translation process
from a specific source human language to other target languages. The
advantage of UniArab lies in the deployment of an interlingua architecture
which uses a robust functional linguistic model founded on RRG in the
machine translation kernel. UniArab is flexible and scalable for multilingual
generation. One of the primary strengths of UniArab is the accurate repre-
sentation of the RRG logical structure of an Arabic sentence. As an inter-
lingua MT application, UniArab has both parse and generate functions
(Nolan and Salem 2010, 2011). Currently, UniArab covers a representative
broad selection of words and can translate intransitive, transitive and

ditransitive clauses, as well as Arabic copular-like nominative clauses. Uni-

Arab has been shown to outperform existing machine translators in the
processing of simple sentences, confirming that RRG is a capable candidate
for interlingua-based machine translation. UniArab can accept as input
individual sentences or whole paragraphs of text consisting of multiple
sentences. An area of future development with the UniArab model is to
(i) address the problem of word sense disambiguation and lexical polysemy,
and (ii) treat complex sentences and nexus–juncture relations.
21.4 Language-Aware Conversational Agents and Avatars
In this section we highlight important computational work, using RRG,

which underpins the development and implementation of (i) text-based
conversational agents and (ii) humanoid avatars for Sign Language.
21.4.1 Speech Acts, Knowledge and Conversational Agents

Following on primarily from the work on machine translation described
earlier that implemented the RRG bidirectional linking system of RRG in
software, recent work has advanced and implemented a model of conversa-
tional agents in a computational framework that builds on the notion of
speech acts from discourse, as a proof of concept (Panesar 2017, 2019).
Intelligent conversational agents are a software paradigm that can be used
to exploit the possibilities presented by human knowledge, human language
and human/agent spoken and written communications, possibly over mas-
sive open distributed systems such as the Internet. In this conversational
agent framework, a view of software-based agents is proposed based on
Nolan (2014). A conversational agent has both ‘internal’ and ‘external’
models to support the speech acts (Searle 1969), common ground formulation
(Kecskes and Zhang 2009) and discourse. The internal model of the agent is
concerned with the internal state of the agent, based upon the intersection
at any given time of the agent’s internal beliefs, desires and intentions,
known as BDI states. The external model of the agent is composed of an
interaction model with its world (human agent and other agent). These
models support the construction of a conversation-specific emergent
common ground. Importantly, the conversational agent also has a lan-
guage model in software that is related to its interaction model to support
the interactive bidirectional communication in human language through
speech acts. The linguistic model employed to motivate the design of the
language model is RRG (Nolan 2012; Van Valin 2005) with its bidirectional
linking system.
The agent framework has been implemented (Panesar 2019, 2020) as a
model of conversational agents in a computational framework. It builds on
the notion of a speech act from discourse within the functional RRG model

800 BRIAN NOLAN
of grammar, and the construction of a common ground in a discourse

workspace to underpin the conversational interaction. The framework con-
nects the intelligent conversational agent paradigm to the RRG model of
language, leveraging the notion of constructions as grammatical objects. It
builds on work on understanding constructions as grammatical objects
within RRG (Nolan 2011, 2012, 2013; Diedrichsen 2011; Nolan and Diedrich-
sen 2013) and the role of computational approaches to functional grammars
(Nolan and Periñán-Pascual 2014). This framework has potential for use in
testing hypotheses to support claims of adequacy (Butler 2009) within an
RRG approach. It extends our understanding of the grammar–pragmatics
interface and the construction of an emergent common ground between
interlocutors. The management framework for a language understanding
system supporting conversational agents is seen in Figure 21.5, with the
applicable sub-models, in particular with the language model based on RRG.
Here, the natural language processing system is at the heart of
the architecture.
To build an intelligent conversational agent in software one needs to
represent a number of important aspects of a conversational agent-based
system. One needs to capture (i) the set of beliefs that the agent has at any
Figure 21.5 Dialogue management framework for a language understanding system supporting conversational agents
and a dynamic common ground

given time, (ii) the goals that the agent will try to achieve, (iii) the actions that
the agent performs, (iv) the knowledge of the effects of these actions, (v) the
environment information the agent has (which may be incomplete or incor-
rect), (vi) the ongoing discourse interaction that the agent has with other
(human) agents and their environment over time, (vii) human language
understanding, and (viii) conversation tracking over a discourse. The parameters
of the conversational agent are characterized in its (i) personality and
emotional status, (ii) awareness of social context, (iii) model of the conversa-
tional partner (¼ other conversational actor), (iv) dialogue tracking, (v) agent
dialogue-turn tracking, and (vi) linguistic information concerning language-
specific grammar and clause structure and speech acts.
In the architecture outlined here, the relationship between agent param-
eters and behavioural choices made during the parse/generation of natural
language conversation is governed by the Agent Dialogue Manager.
Typically, this rule-based system consults other elements of the architecture
to maintain the BDI state of the conversational agent, its personality and
emotional status etc. to control a generalized natural language conversation
system and supply the appropriate parameters for conversational behaviour.
Importantly, a model of an agent system is distinguished by its active use of
social concepts.
Panesar (2017, 2019, 2020) designed and implemented this linguistically
orientated, conversational software agent framework as a proof of concept.
She discusses the relationship between natural language processing and
knowledge representation, and connects this with the goals of RRG
(Figure 21.6). In this implemented framework, Phase 1 is concerned with
Figure 21.6 Conceptual framework of the conversational software agent (from Panesar 2017: 190)

802 BRIAN NOLAN
the RRG language model. Phase 2 manages the agent cognitive model and its
interfaces with the BDI model, planning model and knowledge model, while
Phase 3 handles the agent dialogue model, including dialogue management
via the RRG language model.
The advantages provided by RRG (Panesar 2019) reside in its fitness-for-
purpose for computational implementation and its level of computational
adequacy. Panesar implements a computational model of the linking algo-
rithm that utilizes a speech act construction as a grammatical object and
the (BDI) sub-model of beliefs, desires and intentions (Rao and Georgeff
1995). This work builds on elements of earlier research on the RRG
interlingua-based machine translation engine described earlier. This RRG-
based conversational software agent framework has been successfully imple-
mented in software (Panesar 2017, Pokahr et al. 2014) using conceptual
graphs and the resource description framework (RDF).
The innovation of this research resides in the combination of models and
their interoperability. The research provides new insights into the interface
between language and knowledge, via an RRG perspective.
21.4.2 Modelling Sign Language in an RRG Avatar

Computational modelling using RRG has been applied to a humanoid avatar
to motivate an extended architecture for the lexicon to accommodate the
requirements of visual gestural sign languages (SL). Murtagh (2019a, 2019b)
designed and implemented a signing avatar which used RRG as its linguistic
model. Due to the visual gestural nature of Irish Sign Language (ISL), and
indeed all sign languages, and the fact that ISL has no written or aural form,
to communicate an ISL utterance in computational terms required the
implementation of a humanoid avatar capable of movement within three-
dimensional (3D) space (Figure 21.7). In the creation of the avatar, it was
Figure 21.7 The avatar model (from Murtagh 2019b: 95)

necessary to define all of the ISL phonological parameters in computational

terms. The lexicon extensions involve an enhancement to qualia structures
and qualia theory (Pustejovsky 1991, 1995). For the representation of sign
languages, Murtagh posited an articulatory structure level in a framework
called Sign–A. This research is work in progress in the development of a full
linguistically motivated computational framework for ISL, based on RRG,
and involving the avatar.
In characterizing a sign language, one needs to refer to the various sign
articulators (hands, fingers, eyes, eyebrows, etc.), as these are what are used
to articulate various phonemes, morphemes and lexemes of an utterance
(Figure 21.8). According to Murtagh (2019a), the visual gestural realization of
a word in SL involves the simultaneous and parallel expression of a varied
number of manual features (MFs) and non-manual features (NMFs), together
in concert over a timeline. The MF phonological parameters are defined as
location, orientation, movement and handshape or relative configuration. The NMF
phonological parameters are eyebrow movement, movement of the eyes and
eyelids, mouth patterns, tongue movement, and blowing of the cheeks, and it also
includes head tilting and shoulder movement. This description of the articula-
tors and how they act in concert in an utterance is modelled in this avatar in
RRG linguistic terms and represented in the lexicon. The new level of lexical
representation, the articulatory structure level, is based on an extension of
the qualia theory from the generative lexicon of Pustejovsky (1991, 1995)
and caters specifically for computational linguistic phenomena consistent
with signed languages, enabling the adequate representation of ISL within
the RRG lexicon.
To facilitate the development of this RRG-motivated computational frame-
work for ISL, all of the ISL phonological parameters, for both MFs and NMFs,
are rigorously described and defined in computational terms, taking into
consideration a sign realization timeline, an event duration, and a 3D
signing space. Any given ISL phoneme, morpheme or lexeme may be
Figure 21.8 The hand articulator in 3D space with nodes on fingers and thumb (from Murtagh
2019b: 98)

804 BRIAN NOLAN
realized simultaneously and in parallel along a timeline. The position of the

hands in 3D space has consequences at a syntactic and also at a semantic
level for ISL. Also, morphemes are articulated at particular points or loci in
relation to the signer for pronominal and anaphoric reference. In addition,
the order or linear sequence in which these units are realized is significant
with regard to the syntax and semantics of the language. Therefore, all the
information applicable to the signed lexical items must be accounted for in
order to accurately represent the language. The parameters required to
represent ISL can be explained more easily if one uses the analogy of
instruments playing together in an orchestra, where the various articulators
would be represented by the instruments and each articulator will play its
own part in producing an overall production or articulation, similar to the
instruments playing their parts in an orchestra.
This research by Murtagh (2019a, 2019b) fuses a constructional perspective
of important phenomena in SLs, in particular ISL, within the RRG context.
The innovation within Murtagh’s research is that it leverages the develop-
ment of an avatar within 3D space, employs RRG as the linguistic model,
and provides computational phonological parameters capable of represent-
ing and realizing ISL lexemes and their articulatory structure levels within
the extended architecture of the RRG lexicon. Timeline and event duration
are included, as is a definition of the 3D signing space and the ISL body
anchored parameters. This development, called the Sign–A framework by
Murtagh, has the potential to be employed as a linguistically motivated
application of avatar technology for sign languages, based on implementing
RRG in software, and to be used for real-time text-to-sign machine
translation.
21.5 Functional Grammar Knowledge Base

Environment
21.5.1 The FunGramKB Initiative

FunGramKB2 (‘Functional Grammar Knowledge Base’) is an online envir-
onment for the semi-automatic construction of a multipurpose lexico-
conceptual knowledge base for natural language processing (NLP)
systems and natural language understanding. It is multifunctional and
multilingual, and designed to support many NLP tasks, including infor-
mation retrieval and extraction, machine translation, dialogue-based
systems, and so on, and with many natural languages (Guerra García
and Sacramento Lechado 2014; Felices-Lago and Ureña Gómez-Moreno
2014; Montiel-Ponsoda and Aguado-de-Cea 2014; Sánchez-Cárdenas and
Faber 2014).
Periñán-Pascual and Arcas-Túnez (2010) describe the FunGramKB know-
ledge base as comprising three major knowledge levels, consisting of
several independent but interrelated modules. At the lexical level, the

Lexicon stores morphosyntactic, pragmatic and collocational information

about lexical units. The Morphicon helps the system to handle cases of
inflectional morphology. At the grammatical level, the Grammaticon
stores the constructional schemata which help RRG to construct the
semantics-to-syntax linking algorithm. At the conceptual level, the Fun-
GramKB Ontology is presented as a hierarchical catalogue of the concepts
that a person has in mind. Semantic knowledge is stored in the form of
meaning postulates. It consists of a general-purpose module (Core Ontology)
and several domain-specific terminological modules (Satellite Ontologies).
The Core Ontology is an IS-A conceptual hierarchy which allows non-
monotonic multiple inheritance. The Core Ontology takes the form of a
universal concept taxonomy, where ‘universal’ means that every concept
we imagine has, or can have, an appropriate place in the ontology. Import-
antly, the Core Ontology is linguistically motivated but not language-
dependent. The FunGramKB Cognicon stores procedural knowledge by
means of scripts (i.e. conceptual schemata in which a sequence of stereotyp-
ical actions is organized on the basis of temporal continuity). In the Cogni-
con, scripts are structured into predications within a linear temporal
framework, based on a representation of time as a partially ordered graph
where nodes represent events and arcs are tagged with one or more relations
of temporal ordering. In FunGramKB, every predication included in a script
represents an event E which is treated as an interval consisting of a pair of
time points (i, t) (i.e. the start time-point (i) and the end time-point (t)). The
FunGramKB Onomasticon stores information about instances of entities
and events, and stores two different types of schemata (i.e. snapshots and
stories), as instances can be portrayed synchronically or diachronically. The
Onomasticon stores information about named entities and events (i.e.
instances of concepts), in the form of bio-structures. The population of the
Onomasticon is taking place semi-automatically, through the extraction of
structured information from Wikipedia, to turn this information into a rich
knowledge base of more than 2.6 million entities. This knowledge base is
accessible on the Internet.
In the FunGramKB architecture (Figure 21.9), every lexical or grammatical
module is language-dependent, whereas every conceptual module is shared
by all languages. In other words, a computational linguist must develop one
lexicon, one morphicon and one grammaticon for English, one lexicon, one
morphicon and one grammaticon for Spanish, and so on. Knowledge engin-
eers build just one ontology, one cognicon and one onomasticon to process
any language input conceptually. FunGramKB adopts a conceptualist
approach as the ontology becomes the pivotal module for the whole
architecture.
The FunGramKB lexical and grammatical levels are grounded in the
functional-cognitive linguistic theories of RRG and the related Lexical Con-
structional Model (LCM) (Ruiz de Mendoza and Mairal Usón 2008; Mairal
Usón and Ruiz de Mendoza 2009), allowing the FunGramKB system to

806 BRIAN NOLAN
Figure 21.9 The architecture of FunGramKB (source: www.fungramkb.com)
capture syntactic-semantic generalizations which are able to provide both

explanations and predictions of language phenomena.
They state that, at the FunGramKB conceptual level, the model of a
‘scheme’ is fundamental to the representation of the world knowledge.
In the knowledge base, conceptual schemata are classified according to the
parameters of prototypicality and temporality. Conceptual representations
can store prototypical knowledge or can serve to describe instances of
entities or events. A key factor in successful reasoning within an NLP
system is that all the knowledge schemata must be represented with the
same formal language, so that effective information sharing can take place
across all conceptual modules. In FunGramKB, this formal conceptual
representation language is called COREL. Because the FunGramKB concep-
tual modules use COREL as the common internal language for schemata
representation, natural language understanding systems will only require
one common reasoner.
In RRG, the semantic and the syntactic components are mapped via a
linking algorithm, which includes a set of rules that account for the syntax–
semantics interface. As a result, RRG allows an input text to be represented
in terms of a logical structure. In FunGramKB, the RRG logical structure has
been enhanced by a new formalism called the Conceptual Logical Structure
(CLS) (Periñán-Pascual and Mairal Usón 2009). The benefits are that CLSs are
real language-independent representations, since they are made of concepts
and not words. One of the consequences of this interlingua approach is that
redundancy is minimized while informativeness is maximized. The benefits

of interlingua-based systems in multilingual settings, which we have

already discussed in relation to UniArab, are also highlighted by Periñán-
Pascual and Arcas-Túnez (2014), based on an implementation of RRG in the
FunGramKB software. The inferential power of the reasoning engine is more
robust if predictions are based on cognitive expectations. In order to per-
form some reasoning with the input, the CLS is transduced into a COREL
representation, so that it can be enriched by the knowledge in meaning
postulates, scripts, snapshots and stories.
According to Ruiz de Mendoza and Mairal Usón (2008) and Mairal Usón
and Ruiz de Mendoza (2009), as it is grounded within the RRG framework,
the LCM goes beyond the core grammar and incorporates meaning dimen-
sions that have a long tradition in pragmatics and discourse analysis. The
LCM recognizes four levels of constructional meaning, and in the Gramma-
ticon, each one of these constructional levels is computationally imple-
mented into a Constructicon. Level 1 (argumental layer) provides the core
grammatical properties of lexical items. Level 2 (implicational layer) is
concerned with the inferred meaning related to low-level situational cogni-
tive models (or specific scenarios), which give rise to meaning implications
of the kind that have been traditionally handled as part of pragmatics
through implicature theory. Level 3 (illocutionary layer) deals with trad-
itional illocutionary force, which is considered a matter of high-level situ-
ational models (or generic scenarios). Level 4 (discourse layer) addresses the
discourse aspects, with particular emphasis on cohesion and coherence
phenomena.
The FunGramKB Suite provides a set of user-friendly tools to browse, check
and edit the knowledge base. Specifically, conceptual, lexical and grammat-
ical modules can be browsed via a GUI, displaying specific feature-value
information about their elements. At this point in time, FunGramKB is a
mature multipurpose knowledge base for NLP systems. Its goal is the cre-
ation of an NLP knowledge base sufficiently robust to help language engin-
eers to design intelligent natural language understanding systems. The
FunGramKB conceptual level enables the full integration of semantic, pro-
cedural and episodic knowledge by sharing both the knowledge representa-
tion language and the reasoning engine. As a result, expectations on the
occurrence of typical events in a given situation are based on COREL sche-
mata, a concept-oriented interlingua whose inferential power is greater
than the traditional approach to lexical semantics. The FunGramKB lexico-
grammatical levels are grounded in RRG to capture syntactic–semantic
generalizations which can manage and interpret data. Both the RRG and
the related LCM frameworks inspired the construction of the CLS, a lexically
driven interlingua through which the system is able to predict a wide range
of linguistic phenomena in the language generation process. The CLS serves
as the pivot language between the input text and the COREL representation,
while COREL serves as the pivot language between the CLS and the
automated reasoner.

808 BRIAN NOLAN
21.5.2 FunGramKB and ARTEMIS

Another project which aims to provide a fully implemented computational
model is ARTEMIS (‘Automatically Representing Text Meaning via an
Interlingua-Based System’) within FunGramKB. It exists as an application
within the environment of FunGramKB, implemented with a parser for the
computational treatment of the syntax and semantics of sentences. ARTE-
MIS deploys the FunGramKB knowledge base which exploits constructional
schemata as a system for meaning representation (Periñán-Pascual and
Arcas-Túnez 2004, 2005, 2007, 2008, 2010; Periñán-Pascual and Mairal Usón
2009, 2010, 2011; Periñán-Pascual 2013; Cortés-Rodríguez 2016a, 2016b; Díaz
Galán and Fumero Pérez 2016, 2017; Fumero Pérez and Díaz Galán 2017).
ARTEMIS is an NLP application, with a syntactic parser, designed primarily
for natural language understanding. ARTEMIS is linguistically motivated
and adopts the RRG linguistic theory as a foundation (Fumero Pérez and
Díaz Galán 2017; Díaz Galán and Fumero Pérez 2017). For ARTEMIS, RRG has
a number of features that make it suitable for NLP work.
1. RRG is a model where morphosyntactic structures and grammatical rules

are explained in relation to their semantic and communicative functions.
2. RRG is a monostratal theory where the syntactic and semantic compon-
ents are connected via a bidirectional linking algorithm.
3. RRG is a model that makes strong claims to typological adequacy.
4. RRG has a model of the lexicon connected with the linking system.
ARTEMIS is a linguistically grounded application, from both a semantic

and syntactic perspective. Theoretically, it draws from two functional lin-
guistic theories, those of RRG and the LCM mentioned earlier. ARTEMIS is
able to formalize meaning by representing it as a logical structure, and its
functional approach entails that grammar can only be explained through
the interaction of syntax and semantics, which is achieved via a bidirec-
tional linking algorithm. The LCM allows ARTEMIS to account for compos-
itional meaning, providing a machine-tractable representation of
constructions, the constructional schemata. One important element that
binds FunGramKB with RRG and the LCM is the integration of the LCM’s
constructional templates and RRG’s lexical representations / logical struc-
tures into the knowledge base’s language-independent formalism for text
meaning representation, the CLS.
The architecture of ARTEMIS has been designed around three main com-
ponents. The first of these is the grammar development environment (GDE),
comprising the set of syntactic, constructional and lexical rules necessary
for the parsing of natural language expressions. The GDE integrates two
types of construct: a catalogue of attribute-value matrixes (AVMs) to describe
grammatical units, and a set of production rules (grammatical, lexical and
constructional) to allow it to produce a feature-based grammar. The second
component is the CLS constructor, which produces an initial CLS text
meaning representation. This is essentially an advancement of RRG’s logical

structures. The third component is the COREL scheme builder, which trans-
forms the CLS into the formal FunGramKB representation language (which
we mentioned earlier, called COREL), ultimately arriving at an extended
COREL scheme. Rules in ARTEMIS can be divided into syntactic rules to
account for the generation/recognition of the underlying layered structure
of a clause (LSC), constructional rules which guide the embedding of the
structure of argument-predicate constructions (L1 constructions in the LCM)
and lexical rules to tokenize the abstract features encoded in the LSC by
utilizing the information stored in the FunGramKB Lexicon and Ontology.
The process involved in understanding natural language with the ARTEMIS
tools is summarized in the following workflow:
ARTEMIS: [USER] ➔ text ➔ CLS representation ➔ COREL Scheme ➔ [Reasoner]
Whereas RRG is able to connect the lexical entries of verbs with their
syntactic realizations via the linking algorithm, it cannot satisfactorily
account for constructional meaning. This requirement to cater for construc-
tions motivated some adjustments to the RRG model. ARTEMIS required a
constructionist linguistic model, the LCM, to be able to process this type of
meaning, since, at sentence level, argument constructions can take prece-
dence over core verbal semantics and may alter the argument structure of
the predicate. To account for such a possibility, Periñán-Pascual and Arcas
Túnez (2014) proposed and implemented in ARTEMIS a modification of the
RRG LSC by adding a new CONSTR-L1 node between the clause and the core
nodes. Cortés-Rodríguez and Mairal Usón (2016) further consider that the
addition of this new node entails the redefinition of the original RRG pre-
core slot position as a pre-C-L1 position. The rationale for this change is that
the pre-core slot may encompass not only those core constituents stated by
RRG, fronted and interrogative elements, but also constituents which are
triggered by a construction. Another necessary adjustment to the RRG
model derives from the fact that ARTEMIS shares some characteristics of
unification grammars, in such a way that parsing relies not only on syntac-
tic rules but also on the semantic and grammatical information contained
in the AVMs. A consequence of this is that, whereas in RRG abstract gram-
matical categories such as illocutionary force, aspect or negation are
described in the operator projection, in ARTEMIS these values, and the
function words associated with them, are represented in the form of
feature-bearing matrices, which now belong to the constituent projection.
In this approach, every grammatical category in the LSC has to be method-
ically described by listing the attributes that define it in a feature-bearing
matrix. At the same time, these attributes need their own description in
another AVM, in such a way that, whenever a category, an attribute or a new
part of speech (POS) is introduced, the corresponding AVM must be created
and stored in the catalogue of AVMs in the GDE. The attributes that charac-
terize each of the categories in the AVM, which would belong in the RRG
operator projection, should also in turn be defined by AVMs.

810 BRIAN NOLAN
One of the key differences between the linguistic models adopted by

ARTEMIS and RRG is that, in ARTEMIS, constructional meaning is a defining
feature. Constructional meaning in ARTEMIS can be derived from the infor-
mation contained in the core grammar of the verb available in the Lexicon, in
the form of lexical templates, and from the constructional schemata con-
tained in the Grammaticon. This module compiles the description of con-
structions as AVMs which enumerate the main features and establish the set
of constraints that characterize each construction. Following the LCM, the
Grammaticon stores four types of construction: argumental (L1), implica-
tional (L2), illocutionary (L3) and discursive (L4). To capture the difference
between the information provided by the Lexicon and the Grammaticon,
Periñán-Pascual and Arcas-Túnez (2014: 181–187) make a distinction between
kernel constructions and other argumental constructions. The kernel con-
structions are determined by the semantics of the verb, which, depending on
the variables in the lexical template stored in the Lexicon, can be: Kernel 0
(zero argument verbs), Kernel 1 (intransitive), Kernel 2 (monotransitive) and
Kernel 3 (ditransitive), The second set of constructions are those argumental
constructions which cannot be derived from the lexical templates. ARTEMIS is
the first system to employ a substantial knowledge base to generate a full rich
logical structure for use by NLP applications requiring language comprehen-
sion capabilities (Pascual and Arcas-Túnez 2014: 171).
The research by the FunGramKB team (Mairal Usón and Cortés-Rodríguez
2017) has been concerned with the development of a natural language
processing laboratory using tools developed within the framework of RRG.
As a result, a number of computational resources have been built with a
focus on different domains. ARTEMIS is one of the computational resources
that form part of the research project FUNK Lab, which uses the linguistic
theories of RRG and the LCM. The different domains are:
1. ARTEMIS, Automatically Representing Text Meaning via an Interlingua-

based System, is a computational resource which allows for the auto-
matic generation of a conceptual logical structure, that is, a fully speci-
fied semantic representation of an input text.
2. Navigator, a tool that allows a user to retrieve data from the lexical entries
in the English lexicon (e.g. morphosyntactic, pragmatic and collocational
information) and from the conceptual entries in the core ontology (e.g.
thematic frame, meaning postulate), as developed within the framework of
the FunGramKB project. This resource allows the user to navigate through
the linguistic (the English lexicon) and the conceptual levels (the ontology).
3. RONDA, RecOgniziNg Domains with IATE, is a tool used to categorize a
text or a collection of documents in different specialized domains.
4. CASPER, CAtegory and Sentiment-based Problem FindER, is a resource to
analyse micro-texts (e.g. tweets) for the automatic detection of user-
defined problems by following a symbolic approach to topic categoriza-
tion and sentiment analysis.

5. DAMIEN, DAta MIning Encountered, is a workbench that allows research-

ers to do text analytics by integrating corpus-based processing with stat-
istical analysis and machine-learning models for data-mining tasks.
6. DEXTER, Discovering and Extracting TERminology, is a tool developed as
an online multilingual workbench. It is provided with a suite of tools for
(i) the compilation and management of small and medium-sized corpora,
(ii) the indexation and retrieval of documents, (iii) the elaboration of
queries by means of regular expressions, (iv) the exploration of the
corpus, and (v) the identification and extraction of term candidates (i.e.
unigrams, bigrams and trigrams).
The future direction of the work of the FUNK lab, including ARTEMIS, is to
move towards a knowledge-informed intelligent cognitive agent which has
RRG as the linguistic model. This research is ongoing.
21.6 The RRG Contribution to NLP
The RRG theoretical framework creates a relationship between syntax and

semantics and can account for how semantic representations are mapped
into syntactic representations. There are several properties of RRG as a
theory that make it well suited to computational implementation. First,
the theory has a semi-formal logical formalism. This enables the develop-
ment and maintenance of software systems that interpret the formalism.
Implementing the RRG theory in software requires translating its semi-
formal logical formalism into code. In particular, this translation of the
theory into code poses challenges in the implementation of the lexicon and
the bidirectional linking system of RRG. Furthermore, because the linking
system in RRG is bidirectional, an implemented grammar can be used for
both parsing and generation. As well as its typological adequacy, RRG has a
level of computational adequacy which allows for the creation of algorithms
that are computationally tractable. This enables RRG grammars to be used
in a range of applications, and enables linguists to test hypotheses. The
languages that have undergone some level of computational treatment in
RRG include English, Irish, Modern Standard Arabic, Irish Sign Language,
Biblical Hebrew, German and the Romance languages (Spanish, France,
Italian, Romanian). Sign languages (Murtagh 2019a, 2019b), as visual ges-
tural languages, pose interesting problems for both functional models of
grammar (Leeson and Nolan 2008) and computational models of these
languages. A consequence of this computational work has been the enrich-
ment of the theoretical elements of the RRG theory as a functional model,
especially in its semantic and lexical underpinnings, where they connect
with concepts, and the building of software applications that demonstrate
its viability in natural language processing. This computational work pro-
vides compelling evidence that the functional approach of RRG has a

812 BRIAN NOLAN
positive and crucial role to play in natural language processing. RRG delivers
a credible and realistic linguistic model to underpin the kinds of NLP
applications discussed in this chapter, and also in applications with a
functional typology orientation.
Why should a grammar deal with linking from syntax to semantics at all?
Should not specifying the possible realizations of a particular semantic
representation suffice? The answer of course is ‘no’. Functional linguists of
all persuasions refute this using the argument that theories of linguistic
structure should be directly relatable to testable theories of language pro-
duction and comprehension. Computational models assist in this process.
Based on our experience to date with computational models of RRG, we can
verify that RRG is suitable as a basis for computational implementation in
software. Computational models of RRG serve as a rich testing ground for
the linguistic theory. Indeed, Butler and Gonzálvez-García (2014: 451) note
that RRG is the only functional model in what they call ‘functional-cognitive
space’ that has a number of ongoing computational research activities.
Butler and Gonzálvez-García note in particular several of the case studies
discussed here (Guest 2008; Gottschalk 2012, 2019; Nolan 2004; Nolan and
Salem 2010, 2011; Murtagh 2019a, 2019b). In addition, recent work in
connection with the RRG-inspired FunGramKB knowledge base has seen
important advances in the computational implementation of RRG-based
procedures. Butler and Gonzálvez-García consider these studies as represent-
ing an important growing trend in RRG (see also Butler and Martín Arista
2009). An important question (Nolan 2016), from a computational RRG
perspective is: ‘What can theoretical linguistics do for natural language
processing research?’ In order to be considered fit for purpose, certain levels
of adequacy are expected from contemporary linguistic models (Butler
2009). These levels span the descriptive, typological, psychological and
explanatory adequacy required in a viable model of language. In today’s
world we can consider that another level of adequacy is also required, that
of computational adequacy. This is a concept well known to those scientists
and professionals working within computer science and software develop-
ment. A computer system and its underlying model must be fit for purpose
as specified. It must deliver its results in a coherent, timely and efficient
manner with efficient utilization of available resources. In linguistic model-
ling terms, a linguist is concerned with the model of the computation in the
mind of a person who actually computes the link between the concept–
semantics–syntax interfaces in an utterance. In computational linguistic
terms, however, this model is then a subset of the model of grammar used,
and once implemented in software, its systematic behaviour. Specifically,
considered as a computational system, does the model work to deliver the
correct and expected results optimally and efficiently? An implication of the
requirement of computational adequacy for a linguistic model imple-
mented in software is that one must address the appropriate levels of
granularity required for software specifications in order to actually model

a grammar. This extra level of detail needed to specify the model of gram-
mar such that it is understandable by a computer system puts additional
and complex demands on the linguist to reach the correct levels of granu-
larity and formalized precision so that the model can be programmed.
For RRG, where syntax is not viewed as autonomous but semantically
motivated, the scale of the challenge is increased. However, as a scientific
enterprise, it is worthwhile as it has the potential to model the computa-
tion of language from lexicon, semantics and morphosyntax. In function-
ally motivated models of language a grammar–lexicon–construction
continuum exists and the interfaces between these need to be rigorously
expressed (Nolan and Periñán-Pascual 2014; Nolan and Salem 2011; Nolan
2014). As a way of treating linguistic and computational complexity in
pursuit of linguistic realism with a robust cultural awareness of linguis-
tic conventions, this scientific enterprise of creating linguistically
motived language software and computational models of grammar is
crucial. The combination of computer science with RRG-style linguistics
has the potential to create many useful innovations that will cause a
phase shift in our expectations (Nolan and Diedrichsen 2019; Kailuweit
et al. 2019). As it has become increasingly pervasive in our lives, human
language technology is now a central component of computer science
and computational linguistics in its treatment of natural language in
software. Central to future research and development in this area is the
convergence of knowledge of computer science and linguistics and other
related disciplines. Crucially, the key ingredients are an understanding of
linguistics and how languages work to meet the levels of adequacy
needed to be successful and accepted, and to guide computer science in
application development and the engineering of solutions of value to us.
The functionally oriented RRG theory of linguistics can contribute con-
siderably to NLP research.
One way to measure a syntactic theory in practical computational appli-
cations is to consider the computational complexity of its implemented
parsing and generation algorithms. At its present stage of development,
computational work in RRG has tended not to address issues relating to
computational complexity at a theoretical or empirical level. To date, the
focus of computational work has been on the creation of RRG-based soft-
ware applications that serve some particular purpose. Consequently, com-
putational work on RRG has not focused theoretically on the science of
parsing, concentrating instead on implementing linguistically motivated
parsers for particular applications and a variety of languages. One excep-
tion to this is found in Kallmeyer et al. (2013), which describes a body of
research on the computational formalization of RRG that uses the building
blocks provided by the theory of Tree Adjoining Grammar (TAG), with
modified operations for combining trees. In this, Kallmeyer et al. (2013:
175) report on a tree rewriting system, called Tree Wrapping Grammar
(TWG), that captures the basic tree-composition principles of RRG, with the

814 BRIAN NOLAN
aim of formalizing the composition of syntactic templates and their com-

positional mechanisms, while drawing on concepts from TAG theory. The
authors provide two operations for syntactic composition, which they call
(wrapping) substitution and sister adjunction. The first operation provides
a model of argument insertion as well as the construction of long-distance
dependencies. The second operation implements adjunction to non-binary
trees. Their approach to tree construction and parsing is part of a larger
project that aims at a full formalization of RRG. An advantage of such a
formalization is that it facilitates a computational treatment of the gram-
mar. The authors define a Cocke–Younger–Kasami (CYK) parsing schema
for tree wrapping grammars, which can be employed for RRG parsing. Such
a formalization of the grammatical structures of RRG consists of two parts
(Kallmeyer et al. 2013: 177): the specification of the elementary trees and
the definition of the compositional operations acting on them. Kallmeyer
et al. (2013: 183) provide a formal definition of the elementary trees and
the composition operations (wrapping) substitution and sister adjunction
that operate on them in the form of a TWG. The syntactic inventory of RRG
can be characterized as a formal grammar of this type. A finite set of
elementary trees is specified via a meta-grammar, including the formula-
tion of RRG’s linking theory and the specification of different syntactic
realizations for a particular valence frame. These elementary trees are then
combined into larger trees using the TWG. According to the authors, the
TWG can generate all context-free languages, and cross-serial dependencies
can be described (Kallmeyer et al. 2013: 189). This work contributes to the
formalization of RRG, and, clearly, this is a direction for future research.
Various projects have developed computational resources on the basis of
RRG, in particular the FunGramKB project, which is quite mature at this
stage. In this chapter, we have sketched the landscape of computational
work in RRG. We have discussed how RRG as a theory supports this computa-
tional work, described a number of computational projects that use RRG,
and highlighted some implemented software systems using RRG. These
applications and their target languages are quite diverse. RRG grammars
have, in the main, been used for practical tasks, where the syntactic and
semantic analyses provide useful features. In one important NLP task,
machine translation uses a grammar to parse and also to generate. We
explored ways in which computational work can inform linguistic research
and indicated some areas of future research direction. This field is dynamic
and we hope this chapter will encourage interested linguists to engage in
RRG research in computational linguistics.
References
Butler, C. S. 2009. Criteria of adequacy in functional linguistics. Folia Linguis-

tica: Acta Societas Linguistica Europaea 42(1): 1–66.

Butler, C. S. and F. Gonzálvez-García., F. 2014. Exploring Functional-Cognitive

Space (Studies in Language Companion Series 157). Amsterdam: John
Benjamins.
Butler, C. S. and J. Martín Arista (eds.). 2009. Deconstructing Constructions.
(Studies in Language Companion Series 107). Amsterdam: John
Benjamins.
Copestake, A. 2002. Implementing Typed Feature Structure Grammars (Center for
the Study of Language and Information Publication Lecture Notes). Stan-
ford, CA: CSLI publications.
Cortés-Rodríguez, F. 2016a. Parsing simple clauses within ARTEMIS: The
computational treatment of the layered structure of the clause in Role
and Reference Grammar. In Proceedings of the 34th International Conference of
AESLA, Alicante, 14–16.
Cortés-Rodríguez, F. 2016b. Towards the computational implementation of
Role and Reference Grammar: Rules for the syntactic parsing of RRG
phrasal constituents. Círculo de Lingüística Aplicada a la Comunicación (CLAC).
65: 75–108.
Cortés-Rodríguez, F. and R. Mairal Usón. 2016. Building an RRG computa-
tional grammar. Onomazein 34: 86–117.
Díaz Galán, A. and M. Fumero Pérez. 2016. Developing parsing rules within
ARTEMIS: The case of Do auxiliary insertion. In C. Periñán-Pascual and E.
Mestre-Mestre (eds.), Understanding Meaning and Knowledge Representation:
From Theoretical and Cognitive Linguistics to Natural Language Processing,
Díaz Galán, A. and Fumero Pérez, M. 2017. ARTEMIS: state of the art and
future horizons. Revista de Lenguas para Fines Específicos 23(2): 16–40. https://
ojsspdc.ulpgc.es/ojs/index.php/LFE/article/view/917.
Diedrichsen, E. 2008. Where is the PreCore slot? Mapping the layered struc-
ture of the clause and German sentence topology. In Van Valin (ed.),
203–224.
Diedrichsen, E. 2011. The theoretical importance of constructional schemas
in RRG. In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar,
Diedrichsen, E. 2014. A Role and Reference Grammar parser for German.
In Nolan and Periñán (eds.), 105–142.
Dorr, B. J. 1992. Machine translation divergences: A lexical semantic perspec-
tive. Proceedings of the Second Seminar on Computational Lexical Semantics,
169–183. Toulouse, France.
Dorr, B. J., E. H. Hovy and L. S. Levin. 2006. Machine translation: Interlingual
methods. In Natural Language Processing and Machine Translation Encyclopedia
of Language and Linguistics (2nd ed.). http://citeseerx.ist.psu.edu/viewdoc/
summary?doi¼10.1.1.58.532.
Felices-Lago, A. and P. Ureña Gómez-Moreno. 2014. FunGramKB term
extractor: A tool for building terminological ontologies from specialised
corpora. In Nolan and Periñán-Pascual (eds.), 251–270.

816 BRIAN NOLAN
Fumero Pérez, M. and A. Díaz Galán. 2017. The interaction of parsing rules
and argument-predicate constructions: Implications for the structure of
the Grammaticon in FunGramKB. Revista de Lingüística y Lenguas Aplicadas
12: 33–44.
Gottschalk, J. 2012. The persuasive tutor: A BDI teaching agent with Role and
Reference Grammar language interface – Sustainable design of a conver-
sational agent for language learning. ITB Journal 13(2): 31–50. https://arrow
.tudublin.ie/itbj/vol13/iss2/3.
Gottschalk, J. 2014. Three-place predicates in RRG: A computational
approach. In Nolan and Periñán-Pascual (eds.), 79–104.
Gottschalk, J. 2019. On the application of conceptual graphs in RRG: First
steps towards a functional computational processing model. In R. Kailu-
weit, L. Künkel and E. Staudinger (eds.), Applying and Expanding Role and
Reference Grammar, 325–354. Freiburg: NIHIN Studies. https://freidok.uni-
freiburg.de/data/16830.
Guerra García, F. and E. Sacramento Lechado. 2014. Exploring the thematic-
frame mapping in FunGramKB. In Nolan and Periñán-Pascual
(eds.), 233–250.
Guest, E. 2008. Parsing for Role and Reference Grammar. In Van Valin (ed.),
435–453.
Jackendoff, R. S. 1990. Semantic Structures. Cambridge, MA: MIT Press.
Kailuweit R, L. Künkel and E. Staudinger (eds.). 2019. Applying and Expanding
Role and Reference Grammar. Freiburg: NIHIN Studies. https://freidok.uni-
freiburg.de/data/16830.
Kallmeyer, L., R. Osswald and R. D. Van Valin, Jr. 2013. Tree wrapping for Role
and Reference Grammar. In G. Morrill and M.-J. Nederhof (eds.), Proceedings
of the Formal Grammar 17th and 18th International Conferences, 175–190. Hei-
delberg: Springer.
Kecskes, I. and F. Zhang. 2009. Activating, seeking, and creating common
ground: A socio-cognitive approach. Pragmatics & Cognition 17:(2): 331–355.
Leeson, L. and B. Nolan. 2008. Digital deployment of the Signs of Ireland
Corpus in Elearning. In Language Resources and Evaluation LREC2008 – 3rd
Workshop on the Representation and Processing of Sign Languages: Construction
and Exploitation of Sign Language Corpora, Marrakech, Morocco.
Mairal Usón, R. and F. Cortés-Rodríguez. 2017. Automatically representing
text meaning via an interlingua-based system (ARTEMIS): A further step
towards the computational representation of RRG. Journal of Computer-
Assisted Linguistic Research 1: 61–87.
Mairal Usón, R. and F. Ruiz de Mendoza. 2009. Levels of description and
explanation in meaning construction. In C. S. Butler and J. Martín Arista
(eds.), Deconstructing Constructions (Studies in Language Companion Series
107), 153–198. Amsterdam: John Benjamins.
Montiel-Ponsoda, E. and G. Aguado-de-Cea. 2014. Applying the lexical con-
structional model to ontology building. In Nolan and Periñán-Pascual
(eds.), 313–338.

Murtagh, I. 2019a. Motivating the computational phonological parameters

of an Irish Sign Language avatar. In Nolan and Diedrichsen (eds.), 323–339.
Murtagh, I. 2019b. A Linguistically Motivated Computational Framework for Irish
Sign Language. PhD thesis, Trinity College Dublin School of Linguistic
Speech & Communication Science. www.tara.tcd.ie/handle/2262/89131.
Nolan, B. 2004. First steps towards a computational Role and Reference
Grammar. In Proceedings of the International Conference on Role and Reference
Grammar, Dublin, July 2004, 196–223. https://rrg.caset.buffalo.edu/rrg/
RRG2004%20Book%20of%20Proceedings.pdf.
Nolan, B. 2011. Meaning construction and grammatical inflection in the
layered structure of the Irish word: An RRG account of morphological
constructions. In W. Nakamura (ed.), New Perspectives in Role and Reference
Grammar, 64–103. Newcastle upon Tyne: Cambridge Scholars Publishing.
Nolan, B. 2012. The Syntax of Modern Irish: A Functional Account. Sheffield:
Equinox.
Nolan, B. 2013. Constructions as grammatical objects: A case study of the
prepositional ditransitive construction in Modern Irish. In B. Nolan and E.
Diedrichsen (eds.), Linking Constructions into Functional Linguistics, (Studies in
Language Companion Series 145), 143–178. Amsterdam: John Benjamins.
Nolan, B. 2014. Extending a lexicalist functional grammar through speech
acts, constructions and conversational software agents. In Nolan and
Periñán (eds.), 143–164.
Nolan, B. 2016. What can theoretical linguistics do for natural language
processing research? In C. Periñán-Pascual and E. Mestre Mestre (eds.).
Meaning and Knowledge Representation, 235–248. Newcastle upon Tyne: Cam-
bridge Scholars Publishing.
Nolan, B. and E. Diedrichsen (eds.). 2013. Linking Constructions into Functional
Linguistics: The Role of Constructions in Grammar (Studies in Language Com-
panion Series 145). Amsterdam: John Benjamins.
Nolan, B. and E. Diedrichsen (eds.). 2019. Linguistic Perspectives on the Construc-
tion of Meaning and Knowledge: The Linguistic, Pragmatic, Ontological and Com-
putational Dimensions. Newcastle upon Tyne: Cambridge Scholars
Publishing.
Nolan, B. and C. Periñán-Pascual (eds.). 2014. Language Processing and Gram-
mars: The Role of Functionally Oriented Computational Models (Studies in Lan-
guage Companion Series 150). Amsterdam: John Benjamins.
Nolan, B. and Y. Salem. 2010. UniArab: An RRG Arabic-to-English Machine
Translation Software. In W. Nakamura (ed.), Proceedings of the 10th Inter-
national Conference on Role and Reference Grammar (RRG 2009), 243–270. linguis-
tics.buffalo.edu/people/faculty/vanvalin/rrg/ProceedingsofRRG2009_02.pdf.
Nolan, B. and Y. Salem. 2011. UniArab: RRG Arabic-to-English Machine Trans-
lation. In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar,
Panesar, K. 2017. A Linguistically Centred Text-Based Conversational Software Agent.
Unpublished PhD thesis, Leeds Beckett University. www.researchgate.net/

818 BRIAN NOLAN
publication/319376849_PhD_Thesis_%27A_linguistically_centred_text-
based_conversational_software_agent%27.
Panesar, K. 2019. Functional linguistic based motivations for a conversa-
tional software agent. In Nolan and Diedrichsen (eds.), 340–371.
Panesar, K. 2020. Conversational artificial intelligence: Demystifying statis-
tical vs. linguistic NLP solutions. Journal of Computer-Assisted Linguistic
Research 4: 47–79.
Periñán-Pascual, C. 2013. Towards a model of constructional meaning
for natural language understanding. In Nolan and Diedrichsen (eds.),
205–230.
Periñán-Pascual, C. and F. Arcas-Túnez. 2004. Meaning postulates in a lexico-
conceptual knowledge base. In Proceedings of the 15th International Workshop
on Databases and Expert Systems Applications, 8–42. Los Alamitos, CA: IEEE.
Periñán-Pascual, C. and F. Arcas-Túnez. 2005. Microconceptual-knowledge
spreading in FunGramKB. In Proceedings of the 9th IASTED International
Conference on Artificial Intelligence and Soft Computing, 239–244. Anaheim:
ACTA Press.
Periñán-Pascual, C. and F. Arcas-Túnez. 2007. Deep semantics in an NLP
knowledge base. In Proceedings of the 12th Conference of the Spanish Association
for Artificial Intelligence, Universidad de Salamanca, 279–288.
Periñán-Pascual, C. and F. Arcas-Túnez. 2008. A cognitive approach to qual-
ities for NLP. Procesamiento del Lenguaje Natural 41: 137–144.
Periñán-Pascual, C. and F. Arcas-Túnez. 2010. The architecture of Fun-
GramKB. In Proceedings of the International Conference on Language Resources
and Evaluation, LREC 2010, Malta, 2667–2674. www.lrec-conf.org/proceed
ings/lrec2010/summaries/284.html.
Periñán-Pascual, C. and F. Arcas-Túnez. 2014. The implementation of the Fun-
GramKB CLS Constructor in ARTEMIS. In Nolan and Periñán-Pascual (eds.),
165–196.
Periñán-Pascual, C. and R. Mairal Usón. 2009. Bringing Role and Reference
Grammar to natural language understanding. Procesamiento del Lenguaje
Natural 43: 265–273.
Periñán-Pascual, C. and R. Mairal Usón. 2010. Enhancing UniArab with
FunGramKB. Procesamiento del Lenguaje Natural 44: 19–26.
Periñán-Pascual, C. and R. Mairal Usón. 2011. The COHERENT methodology
in FunGramKB. Onomázein 24: 13–33.
Pokahr, A., L. Braubach, C. Haubeck and J. Ladiges. 2014. Programming BDI
agents with pure Java. In German Conference on Multiagent System Technologies,
MATES2014, 216–233. Switzerland: Springer International Publishing.
Pustejovsky, J. 1991. The generative lexicon. Computational Linguistics 17(4):
409–441.
Pustejovsky, J. 1995. The Generative Lexicon. Cambridge, MA: MIT Press.
Rao, A. S. and M. P. Georgeff. 1995. BDI agents: From theory to practice.
In Proceedings of the 1st International Conference on Multi-Agent Systems ICMAS-
95, 312–319.

Ruiz de Mendoza, F. and R. Mairal Usón. 2008. Levels of description and

constraining factors in meaning construction: an introduction to the
Lexical Constructional Model. Folia Linguistica 42(2): 355–400.
Sánchez-Cárdenas, B. and P. Faber. 2014. A functional and constructional
approach for specialized knowledge resources. In Nolan and Periñán-Pascual
(eds.), 297–312.
Searle, J. 1969. Speech Acts. Cambridge: Cambridge University Press.
Van Valin, R. D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge:
Van Valin, R. D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics–Pragmatics
Interface. (Studies in Language Companion Series 105). Amsterdam: John
Benjamins.
Van Valin, R. D., Jr. and E. Diedrichsen. 2006. A Bonsai Grammar for German.
www.academia.edu/26126987/A_Bonsai_Grammar_for_German.
Vauquois, B. 1968. A survey of formal grammars and algorithms for recogni-
tion and transformation in machine translation. Proceedings of the IFIP
Congress-6, 254–260.
Winther-Nielsen, N. 2008. A Role-Lexical Module (RLM) for Biblical Hebrew:
A mapping tool for RRG and WordNet. In Van Valin (ed.), 455–478.
Notes
1 Available from SIL: https://software.sil.org/toolbox/.

2 www.fungramkb.com.

Part Five
Grammatical Sketches

22
A Grammatical Sketch
of Cheyenne (Plains
Algonquian, USA)
Avelino Corral Esteban
Abbreviations

1 first singular IF illocutionary force

2 second singular II inanimate plural
3 third (proximate) singular IMPER impersonal construction
4 fourth (obviative) singular IN initial stem
12 first person plural inclusive INAN inanimate
A actor INFR inferential or dubitative
mode
AN animate LS logical structure
ANA anaphoric MOD modality
ASP aspect N nominal stem
CISL cislocative NMR non-macrorole
CLM clause linkage marker OBV obviative
CNJ Conjunct order PoDP post-detached position
CON conjectural mode PRET preterite or narrative mode
DEIC deictic PrCS pre-core slot
DIR directional PrDP pre-detached position
DITR ditransitive PSA privileged syntactic
argument
ECS extra-core slot REP reportative or attributive
mode
END endophoric RP reference phrase
EQ event quantification TEMP temporal
EPEN epenthetic TNS tense
EVID evidentiality TRNS translocative
EXO exophoric U undergoer

824 AVELINO CORRAL ESTEBAN
FAI animate intransitive final VAI animate intransitive verb

stem
FII inanimate intransitive final VII inanimate intransitive
stem verb
FTA animate transitive final VTA animate transitive verb
stem
FTI inanimate transitive final VTI transitive inanimate verb
stem
I inanimate singular X unspecified actor
22.1 Introduction
This chapter examines the grammar of Cheyenne or Tsėhésenėstsestȯtse (Plains

Algonquian, USA) within the framework of Role and Reference Grammar
(henceforth RRG) (Van Valin and LaPolla 1997; Van Valin 2005).* Although
this grammatical sketch aims to cover as much ground as possible, for
reasons of brevity, it only covers in some depth the interaction of syntax,
semantics, and pragmatics in simple sentences. Thus, by way of illustrating
the pervasive influence of pragmatics in all areas of Cheyenne grammar, the
last part of the paper includes an account of the relation between word
order and information structure, and the influence of pragmatics on mor-
phosyntactic coding and macrorole assignment.
The organization of the chapter is as follows: after a brief account of the
principal morphosyntactic features of Cheyenne, Section 22.2 offers a descrip-
tion of a number of grammatical areas which benefit from analysis in RRG
terms. Section 22.3 provides an account of the relationship between infor-
mation structure and two linguistic mechanisms which are particularly
typical of Algonquian languages, namely the proximate/obviation distinction
and the direct/inverse system. These serve to mediate between macrorole
assignment and morphosyntactic coding and, consequently, compensate for
the lack of a rigid word order and a proper case-marking system in Cheyenne.
22.2 Grammatical Sketch
After introducing a number of key features of Cheyenne grammar (22.2.1), in

this section we discuss basic clause types (22.2.2), verb valence and transitiv-
ity (22.2.3), the marking of core arguments, argument-adjuncts and adjuncts
(22.2.4, 22.2.5) and the linking algorithm (22.2.6).
* The data come mainly from my native consultants, supplemented by existing language materials, such as two collections of
texts (Leman 1980a and 1987) and a dictionary (Fisher et al. 2006). I wish to express my gratitude to the Cheyenne people,
especially my late friend Ralph Redfox, for kindly sharing their knowledge of this language with me. Needless to say, all
errors remain my sole responsibility.

A Grammatical Sketch of Cheyenne 825
22.2.1 Basic Morphosyntactic Properties of Cheyenne

Cheyenne is a polysynthetic language which, like other Algonquian lan-
guages, is believed to exhibit a pragmatically influenced word order (Leman
1999: 37), which is concerned with the way in which the message is formu-
lated, rather than with the identification of semantic roles or syntactic
functions. In accordance with the concept of ‘newsworthiness’ (Mithun
1987), which appears to imply importance and unpredictability, the lan-
guage tends to place any elements expressing relevant information preverb-
ally and any elements expressing trivial information postverbally.
Owing to the pronominal-argument and head-marking nature of Chey-
enne, verbs have an extremely complex templatic structure, which not
only includes bound forms covering the grammatical information of the
core arguments of the predicate, but also derivational stems and par-
ticles expressing grammatical categories like aspect, tense, modality,
etc., which are treated as operators modifying the different layers of
the clause, as shown in Figure 22.1 (modified from Corral Esteban
2017: 310).
The relative ordering of the elements within the verbal complex is as
follows. The person cross-referenced on the verbal prefix of a main clause
always refers to the most pragmatically salient participant, in accordance
with the Person Hierarchy 2 > 1 > 3 > 4 > I(inanimate).1
The preverbal operators between the prefix and the verb stem express the
realis/irrealis distinction, past or future tense, epistemic modality, event
quantification, negation, deontic modality, direction,2 and the different
aspectual distinctions.
Predicates can be simple or complex, depending on whether they com-
prise one single stem or more than one stem. The morphemes constituting a
complex predicate, which incorporate adverbial, nominal and verbal mean-
ings into the verb, are commonly referred to as initial, medial and final, in
accordance with their relative position within the verbal complex (Bloom-
field 1946).
The theme marker indicates voice or, more accurately in Algonquian
terminology, whether the construction is direct or inverse. In broad terms,
the direct construction is used when the subject of the transitive clause
outranks the object in the Person Hierarchy, and the inverse is used when
the object outranks the subject.
Following the theme marker – and sometimes merged with it in a port-
manteau – a number of suffixes that function as pronominal affixes appear,
VERB STRUCTURE - INDEPENDENT ORDER
Prefix Preverb Predicate Theme marker + Suffix(es) Postverb
Salience IRR TNS DIR (2) E. MOD EQ *NEG D. MOD ASP DIR (1) initial+medial+final Voice *NEG Cross-reference EVID IF
Person Number Animacy Prox / Obv
Figure 22.1 Verb structure in Cheyenne (modified from Corral Esteban 2017: 310)

as they cross-reference the core arguments of the predicate and cover gram-
matical information about them in terms of person, number, animacy, and
proximate/obviative status.
The two outermost operators, the clausal operators of evidentiality and
illocutionary force, follow the pronominal suffixes. To summarize, in Chey-
enne operators realized as bound morphemes occur on both sides of the
nucleus.
(1) (tns - neg - mod - asp - dir - V - neg - evid - if)

É - s - sáa - tónėše – éne - e’e-óo’ó’t(a)-ȯ-hé-nȯse-Ø
3-pst-neg-be.able.to-stop-up.in-look.fti-3>I.3sg.a.Isg.u-neg-
evid.3sg.a.Isg.u.rep-decl
ho’honáevose.
mountain
‘He is said to have not been able to stop looking up the mountain.’
Example (1) and its representation in Figure 22.2 indicate that, in Chey-
enne, the order of the morphemes expressing operators with respect to the
nucleus reflects their relative scope and, consequently, confirms the validity
of the principle governing this ordering (Van Valin and LaPolla 1997: 49–52),
SENTENCE
CLAUSE
CORE
PRO NUC PRO RP
PRED
Éi - s – sáa – tónėše – éne - eé - óo´ ´t- ói+j - hé -nósei+j - Ø ho´honáevosej
DIR NUC
ASP NUC
MOD CORE
NEG CORE NEG
TNS CLAUSE
CLAUSE EVID
CLAUSE IF
SENTENCE
Figure 22.2 Layered structure of the clause with constituent and operator projections

since core operators such as internal negation,3 and (deontic) modality

occur closer to the verb than two clausal operators like tense and evidenti-
ality, but are positioned further from the verb than are nuclear operators,
such as aspect or directionals.
Likewise, example (1) shows that the verb in Cheyenne exhibits pronom-
inal affixes, in the form of a prefix such as é- ‘3’, and suffixes, namely -ȯ- and
-nȯse, which cross-reference the different arguments in the clause. Thus, all
the grammatical relations are coded in the verb, rather than in the reference
phrases (RPs), which occur freely in the clause (Corral Esteban 2014: 381).
The person cross-referenced by the prefix refers to the most pragmatically
salient direct core argument in terms of the Person Hierarchy 2 > 1 > 3 >
4 > I, which gives preference to local participants over animate non-local
participants, and to animate non-local participants over inanimate entities.
In intransitive constructions, the prefix always cross-references the only
argument of the verb. However, in transitive constructions, a choice is made
regarding which of the arguments of the verb is more pragmatically salient
in terms of the aforementioned ranking. Furthermore, in transitive predica-
tions, the Person Hierarchy works closely with a Semantic Function Hier-
archy based on animacy and topicality, whereby agent outranks recipient
and benefactive, and recipient and benefactive outrank patient and theme.
A harmonic alignment between the two hierarchies is said to reflect a direct
construction, whereas a disharmonic alignment between them is said to
exhibit an inverse construction.
(2) Né-méhót-o. ! Direct construction

2-love.vta-2>3.2sg.a.3sg.u 4
‘You love her.’
love′ (2sg, 3sg.f)
(3) Né-méhót-a. ! Inverse construction

2-love.vta-3>2.3sg.a.2sg.u
‘She loves you.’
love′ (3sg.f, 2sg)
In these two transitive constructions, the verb has two semantic arguments –
a second-person singular argument and a third-person singular argument.
As second person outranks third person in the Person Hierarchy, both
constructions select the prefix né- ‘2’, regardless of the semantic function
of the argument in each construction. However, despite having the same
prefix, each construction exhibits different relations between the verb and
its arguments, as indicated by the suffixes -o and -a, which reflect a direct
and an inverse construction respectively. In (2), the person cross-referenced
by the prefix né- indicating a second-person participant, corresponds to the
higher semantic macrorole, namely the actor, so the Person and Semantic
Function hierarchies are in proper alignment, and the construction is then
marked as direct. In (3), by contrast, the hierarchies are not properly aligned,

as the same person prefix né- corresponds to the lower semantic macrorole,
that is to say, the undergoer, so the construction is marked as inverse.
22.2.2 Basic Clause Types

Cheyenne, like the other Algonquian languages, exhibits three major div-
isions of verb forms, commonly named ‘orders’ following Bloomfield (1946:
97–103): Independent, Conjunct and Imperative. The Independent order
includes all verb forms other than imperatives that can stand alone, and
the Conjunct order is used for all dependent verb forms. Hence, while the
former generally occurs in main clauses (4), the latter appears in subordin-
ate clauses (5–9).
(4) Hetane é-mane-Ø ma’xemené-mahpe.

man 3-drink.vti-3>I.3sg.a.Isg.u apple-water
‘That man drank the apple juice.’
[do′ (hetane, [drink′ (hetane, ma’xemené-mahpe)]) & ingr consumed′
(ma’xemené-mahpe)]
(5) [. . .] tsé-x-heše-manė-se ma’xemené-mahpe.

cnj-pst-thus-drink.vti-3>I.3sg.a.Isg.u.ind apple-water
‘[. . .] that he drank apple juice.’
(6) [. . .] tsé-’-éše-manė-se ma’xemené-mahpe.

cnj-pst-already-drink.vti-3>I.3sg.a.Isg.u.ind apple-water
‘[. . .] after he drank apple juice.’
(7) [. . .] tsé-x-homá’xe-manė-se ma’xemené-mahpe.

cnj-pst-because-drink.vti-3>I.3sg.a.Isg.u.ind apple-water
‘[. . .] because he drank apple juice.’
(8) [. . .] vé’-manė-stse ma’xemené-mahpe.

cnj-drink.vti-3>I.3sg.a.Isg.u.sbjv apple-water
‘[. . .] if he drank apple juice.’
(9) [. . .] momȯxe-manė-stse ma’xemené-mahpe.

cnj-drink.vti-3>I.3sg.a.Isg.u.sbjv apple-water
‘[. . .] I wish he drank apple juice.’
The main difference between the two verbal orders lies in the function of
their prefix: whereas in the Independent order the prefix serves to express
the most pragmatically salient person, in the Conjunct order it indicates the
grammatical mood, namely realis (e.g. indicative tsé- (5–7)) or irrealis (e.g.
conditional vé’- (8), optative momȯxe- (9), etc.). The prefix in a dependent
clause functions as a clause linkage marker and is usually accompanied by
a preverbal particle that identifies the type of subordinate clause (e.g. -heše-
in complement clauses (5), -éše- in temporal adverbial clauses (6), -homá’xe- in
adverbial clauses of reason (7), etc.), which can lead to an instance of either
daughter subordination (5) or ad-subordination (6–8).

The RRG notion of clause structure is based on two fundamental contrasts:

the contrast between predicating and non-predicating elements, and the
one between those non-predicating elements that function as arguments
and those that do not. Regarding the first contrast, Cheyenne sentences have
two major types of predicate, namely verbal (10) and non-verbal, which can
be nominal (11 and 12) or adjectival (13).5
(10) Néhe hetane é-ónėstán-a he’nétoo’o.

deic.end.prox.an man 3-open.vti-3>I.1sg.a.Isg.u door
‘This man opened the door.’
[do′ (néhe hetane, Ø)] cause [become opened′ (he’nétoo’o)]
(11) Tá’tóhe ma’háéso é-ma’heónė-hetane-(é)ve-Ø.

deic.exo.dist.an old.man 3-holy-man.n-be.fai-3sg.a
‘That old man is a holy man.’
be′ (tá’tóhe ma’háéso, [ma′heónėhetane′ ])
(12) Muhammad Ali é-cassiusclay-h-eve-Ø.

Muhammad Ali 3-Cassius.Clay.in-epen-fai-3sg.u
‘Muhammad Ali is Cassius Clay.’
equate′ (Muhammad Ali, Cassius Clay)
(13) Na-né’ame é-pėhév-ȧhe-o’o

1.poss-parents.pl 3-good.in-fai-3pl.a
‘My parents are good.’
be′ (na-né’ame, [good′ ])
Unlike a verbal predicate, such as ónėstán ‘open’ in (10), the formation of a

non-verbal predicate requires an initial nominal or adjectival stem and a
final verb stem, which functions as a copula. This is evidenced by
ma’heónėhetane ‘holy man’ and -eve ‘be’ in the identificational construction
in (11), Cassius Clay and -eve in the equational construction in (12), and pėhév-
‘good’ and -ȧhe ‘be’ in the attributive construction (13).6
The final stem, by contrast, does not occur in specificational predications,
which are expressed in Cheyenne through a different syntactic construction
involving the presence of a relative clause.
(14) Naa’háanéhe hetane tsé-ama’én-o

deic.ana.dist.an man clm-drive.vti-3>I.3.a-I.u
heóve-amȧho’hestȯtse.
yellow-car
‘That man is the schoolbus driver.’ (lit. ‘That man is the one who drives the
yellow car.’)
be′(naa’háanéhe hetane, [drive′ (naa’háanéhe hetane, heóve-amȧho’hestȯtse)])
The post-copular element in a specificational construction is realized syn-

tactically through a relative clause, such as tsé-ama’én-o heóve-amȧho’hestȯtse
‘who drives the yellow bus’ in (14). This relative clause includes a lexical
verb, so it is able to express more easily the second participant of a

specificational predication, as this is the more referential element of the

construction, unlike in a predicational sentence where the second partici-
pant is less referential or not referential at all.
22.2.3 Verb Valence and Transitivity

Verbs in Cheyenne are divided into four types according to the criteria of
syntactic transitivity, based on the number of direct core arguments, and
animacy: inanimate subject þ intransitive verb (II), animate subject þ
intransitive verb (AI), animate subject þ transitive verb þ inanimate object
(TI), and animate subject þ transitive verb þ animate object (TA).7
The number of macroroles of a verb corresponds closely to its character-
ization according to the notion of syntactic transitivity in Cheyenne, as the
number of direct core arguments is generally the same as that of macro-
roles, provided the arguments refer to a particular entity. Thus, single
macrorole verbs are usually intransitive:
(15) Peter he-mėšemo é-máse-am(e)-ėhn(e)-óho.

Peter 3.poss-grandfather.obv 4-happily-asp.in-walk.fai-4sg.a
‘Peter’s grandfather is walking happily.’
do′ (Peter he-mėšemo, [walk′ (Peter he-mėšemo)])
The activity verb ēhne ‘walk’ occurs in an intransitive construction includ-

ing an animate participant that is lexically realized by the obviative RP
argument Peter he-mėšemo ‘Peter’s grandfather’. This argument appears in
the logical structure (LS) of the verb and is cross-referenced on the verb
through the prefix é- ‘4’ and the pronominal affix óho, which refers to a
fourth person / obviative singular animate patient. As this single direct
core argument of each verb is referential, the verb can be considered to
have one macrorole.
An exception to such a correlation between the number of direct core
arguments and macroroles in intransitive constructions is provided in Chey-
enne, like in other languages, by ‘weather’ verbs.
(16) É-ho’ééto-Ø.
I-snow.vii-Isg.a
‘It is snowing. (lit. ‘It snows.’) snow′(Ø)
The direct core argument of an II verb such as ho’ééto ‘snow’ is cross-

referenced on the verb by the prefix é- ‘I’ and a null pronominal suffix if
their argument is singular, or the suffix -nėstse if it is plural. This is not
actually a semantic argument of the predicate (Van Valin 2008: 8). Therefore,
even though these verbs are classified as intransitive in syntactic terms,
semantically speaking they are atransitive.
There are other intransitive constructions showing a mismatch between
the syntactic valency of a verb and its semantic transitivity, owing to the
influence of referentiality on Cheyenne.

(17) É-h-ma’xe-mané-stove taa’é-va.

3-tns-much.drink.vai-imper night-temp
‘There was a lot of drinking at night.’
do′ (Ø, [drink′ (Ø, Ø)])
This impersonal construction in (17) exhibits an intransitive verb stem: mane

‘drink’. This, along with the presence of the prefix é- ‘3’, implies the presence
of an animate argument. However, the occurrence of an impersonal suffix
such as -htove/stove/nove in this construction does not allow the addition of a
pronominal affix cross-referencing the single argument of the verb, which
prevents the prefix from being referential. Thus, the only direct core argu-
ment of the verb – only coded in the prefix – is not a macrorole in an
impersonal construction, and the verb must be considered atransitive,
which shows non-identity between S-transitivity and M-transitivity.
The fact that Cheyenne only cross-references referential arguments can
also be evidenced in intransitive verbs corresponding to a two-place predi-
cate whose second argument is only implied:
(18) Ná-mané-me.
1-drink.vai-1pl.a
‘We drink.’
do′ (1pl, [drink′ (1pl, Ø)])
The activity verb mane ‘drink’ in (18) has one direct core argument that is coded
on both the prefix ná- ‘1’ and the pronominal affix -me and cross-references a
first-person plural animate argument, namely the first argument in the LS.
However, no reference is made to the second semantic argument of the verb;
hence the verb mane ‘drink’ in (18) can be said to take one macrorole, which is
assigned to the only direct core argument of the verb.
Arguments with generic reference are generally cross-referenced on the
verb in Cheyenne.
(19) Vé’hó’e é-ohkė-šénetam-ovo xamaevo’ėstan(e)-oho.

white.people 3-eq-loathe.vta-3>4.3pl.a.4sg.u Indian.obv
‘White people usually find Indians repulsive.’ (Fisher et al. 2006: 257)
loathe′ (vé’hó’e, xamaevo’ėstanoho)
In this example, both the prefix é- ‘3’ and the pronominal affix -ovo cross-
reference two arguments – a third-person/proximate plural animate argu-
ment and a fourth-person/obviative singular argument. However, despite
being cross-referenced on the verb, the two arguments are not macroroles
since they are non-specific.
There is, however, a context where an argument with a generic reference
is not coded on the verb in Cheyenne. This occurs when there is a close
semantic relationship between the verb and the argument, which would be
a typical object of the verb. In such cases, the object RP is incorporated into
the verb.

(20) a. É-o’ene-mēna-Ø.
3-pick.in-berry.fai-3sg.a
‘He is berry picking.’ (Fisher et al. 2006: 215)
do′ (3sgM, [pick′ (3sgM, o’ene)])
b. É-o’en-anȯtse menȯtse.
(3) -pick.vti-3>I.3sg.a.Ipl.u berry.pl
‘He is picking the berries.’
do′ (3sgM, [pick′ (3sgM, menȯtse)]) & ingr picked′ (menȯtse)
The difference between activities and their telic version, that is to say active
accomplishments, is observed syntactically in Cheyenne. Unlike the active
accomplishment transitive verb o’en ‘pick’ in (20b), the activity intransitive
verb o’enemēna ‘berry-pick’ in (20a) is formed by the initial stem of the verb
‘pick’, o’ene, and a final stem of the noun ‘berry’, mene. Also, while o’en ‘pick’
has two direct core arguments, cross-referenced by the prefix é- ‘3’ and the
suffix -anȯtse, which indexes a third-person singular animate agent and a
plural inanimate patient, the bipartite verb stem o’ene-mēna ‘berry-pick’ has
only one direct core argument – a third-person singular animate agent, which
is coded by the prefix and the null suffix. Additionally, as only the first
argument of the activity verb in (20a) is specific, the verb only takes one
macrorole. Conversely, the active accomplishment verb in (20b) has two
specific arguments and, therefore, takes two macroroles.
Except for these examples showing the influence of referentiality, the
number of direct core arguments and macroroles also tends to coincide in
Cheyenne transitive constructions provided the arguments are referential.
Consequently, two macrorole verbs are usually transitive.
(21) Ná-ono’átam-oo’o tsé-héstoo’éšeeh-ae-se.

1-respect.vta-1>3.1sg.a.3pl.u cnj-raise.vta-3>1.3pl.a-1sg.u.ptcp
‘I respect my parents.’ (lit. ‘I respect who raised me.’)
respect′ (1sg, tsé-héstoo’éšeehaese)
The two-place verb ono’átam ‘respect’ has two arguments, namely a first-
person singular animate experiencer and a third-person plural animate
theme, which are cross-referenced on the verb by the prefix ná- ‘1’, the theme
marker -o- and the pronominal affix -o’o. The second argument is also
realized by a full lexical RP, namely tsé-héstoo’ éšeehaese ‘my parents’ (lit.
‘those who raised me’). As both direct core arguments are referential, the
verb can be said to have two macroroles.
Finally, unlike other Algonquian languages (e.g. Blackfoot, Plains Cree or
Ojibwa), Cheyenne ditransitive verbs have a very complex system of suffixes
(Corral Esteban 2014: 419–422), which vary with respect to the animacy and
number of the theme argument. Thus, verbs such as the stems mét ‘give’,
véestomev ‘ask for’, nomáhtseh ‘steal’, véstomev ‘promise’, or vóo’seh ‘show’
behave like true ditransitive verbs, since they cross-reference the three core
arguments morphologically.8

(22) Né-nomáhtsėh-aenóvo mo’éhno’hāme.

2-steal.ditr-3>2>4.3pl.a.2sg.u.4.nmr horse.obv
‘They stole the horse from you.’
[do′ (3pl, Ø)] cause [become not have′ (2sg, mo’éhno’hāme) & become have′
(3pl, mo’éhno’hāme)].
The predicate nomáhtsėh ‘steal’ in (22) shows an instance of causative accom-

plishment where an activity is carried out by a third-person plural animate
actor, cross-referenced by the suffix -aenóvo. This activity leads to a change
of possession whereby the former participant manages to appropriate
something, which is represented by the fourth-person/obviative participant
mo’éhno’hāme ‘horse’, which previously belonged to the addressee. These three
arguments are coded on the prefix né- ‘2’ and the pronominal suffix -aenóvo.
The influence of referentiality on the cross-reference of core arguments in
Cheyenne (Corral Esteban 2014: 409–410) can also be observed in ditransitive
verbs not including a reference to some of their semantic arguments. In
such cases, the arguments are not cross-referenced on the verb morphology,
which means that both the syntactic transitivity of the verb and the number
of macroroles is reduced:
(23) Ná-mét-onóneo’o nėstámane

1-give.ditr-1>3>I.1pl.a.3pl.u.Isg.nmr 12.poss.food
na-nésonėhaneo′o.
our.children
‘We give our children our food.’
[do′ (1pl, Ø)] cause [become have′ (nanésonėhaneo’o, nėstámane)]
(24) Ná-méa(’tov)-noneo′o na-nésonėhaneo′o.

1-give.vta-1>3.1pl.a.3pl.u 12.poss.children
‘We give away our children to somebody.’
[do′ (1pl, Ø)] cause [become have′ (Ø, nanésonėhaneo’o)]
(25) Ná-mé(a’e)-ánóne nėstámane.

1-give.vti-1>I.1pl.a.Isg.u 12.poss.food
‘We give our food to somebody.’
[do′ (1pl, Ø)] cause [become have′ (Ø, nėstámane)]
(26) Ná-méa(‘e)-ēme.
1-give.vai-1pl.a
‘We give something to somebody.’
[do′ (1pl, Ø)] cause [become have′ (Ø, Ø)]
Although the three-place predicate ‘give’ requires three semantic argu-

ments, in Cheyenne these are only coded on the verb when they refer to
specific participants. The pronominal affixes -o-nó-ne-o’o in (23) cross-
reference the three semantic arguments of the logical structure, namely a
first-person plural agent, a third-person plural recipient nanésonėhaneo’o ‘our
children’, and an inanimate singular theme nȧhtámáne ‘our food’. However,

the pronominal affixes -no-ne-o’o in (24) code only a first-person plural agent
and a third-person plural theme nanésonėhaneo’o ‘our children’; the pronom-
inal affixes -á-nó-ne in (25) code only a first-person plural agent and the
inanimate theme nėstámane ‘our food’; and finally, the pronominal affix -ēme
in (26) only cross-references the first-person plural agent.
The role of referentiality is not only noticed on the pronominal affixes, but
also in the verb stem, which adopts different forms depending on the
number of macroroles the verb takes. Thus, in Cheyenne, the predicate ‘give’
can be expressed through a ditransitive stem mét, a TA stem méa’tov, a TI
stem méa’e, or an AI stem méa’e (23–26). As regards the correlation between
the number of direct core arguments and macroroles, in the intransitive
verb in (26) the only direct core argument of the verb is a macrorole. The
same correlation holds in (24) and (25), as the two transitive verbs have two
direct core arguments and both are macroroles. Since only two macroroles
are allowed in RRG, in the ditransitive construction in (23) only two of the
direct core arguments are macroroles – the first-person plural agent and the
recipient nanésonėhaneo’o ‘our children’, the other – the theme nėstámane ‘our
food’ – being the non-macrorole core argument. They are all identified
through cross-referencing.9
To summarize, except for cases such as constructions including
‘weather’ verbs10 and impersonal constructions, and the problem posed
by generic reference, which can be coded on the verb via either pronom-
inal affixes or noun incorporation, M-transitivity closely corresponds to S-
transitivity in Cheyenne, due to the effect of referentiality on the coding of
arguments and, most importantly, on the form of the syntactic core tem-
plate: the presence of a non-referential argument in a transitive construc-
tion affects not only the pronominal affixes but also the form of the verb
stem, since the non-referential argument is not cross-referenced on the
verb and is, therefore, not represented within the core. Finally, the Chey-
enne examples also show that the number of macroroles that a verb takes
is less than or equal to the number of semantic arguments in its logical
structure.11
A particularly challenging case is a typical Algonquian transitive construc-
tion, commonly referred to as the ‘unspecified actor’ construction (Hockett
1996), as it includes no reference to the agent of the action denoted by
the verb.
(27) Ná-éno’eéh-án-e na-vénótse.

1-heal.vta-X>1–1sg.u 1.poss-tepee.loc
‘I was healed at home.’ (lit. ‘Someone healed me in my tepee.’)
be-at′(home, [do′(Ø, Ø)] cause [become healed′ (1sg, Ø)]])
The two-place predicate éno’eéh ‘heal’ occurs in a transitive construction

including an unspecified argument that corresponds to the first argument
in the LS. Bearing in mind that non-referential participants are not coded

on the verb in Cheyenne, it is not surprising that the unspecified agent in

this construction is not cross-referenced on the verb. This assumption
appears to be supported by the fact that, while the verb stem is transitive,
the suffixal morphology resembles that of the AI paradigm. Furthermore,
the presence of a lexicalized agent in this construction would entail
changing the form of the suffixal morphology. However, we might come
to a different conclusion if we take into account that this construction uses
a different theme marker (e.g. -án or -Ø) depending on whether the speci-
fied argument is local or non-local. A possible explanation for this could be
that the theme markers in this construction indicate direct and inverse
morphology, which means that these bound markers cross-reference an
unspecified participant.12 In summary, it is difficult to gauge whether this
construction reveals an identity between S-transitivity and M-transitivity
or not.
Finally, the transitivity of the verb shows an important difference between
the unspecified actor construction and the impersonal construction illus-
trated by example (17), for, unlike the impersonal construction, which is
built on an intransitive verb stem and eliminates any cross-reference to the
single argument on the verb suffix, the unspecified actor construction
involves a transitive verb but AI morphology, which implies that at least
one argument is coded on the verb.
22.2.4 Argument Coding on the Verb

Only two different types of argument can be distinguished in Cheyenne:
direct core arguments and argument-adjuncts, as core arguments can be
direct but not oblique in this language. Direct core arguments must be
referential to be cross-referenced on the verb. However, if they are third-
person RPs, they can be morphologically marked or unmarked, since Chey-
enne uses a reference-tracking system to establish a clausal disjoint refer-
ence between multiple third-person participants. Thus, in situations
involving more than one third-person RP, the most pragmatically salient
RP (i.e. the proximate) is left unmarked, and any other less pragmatically
salient RP is marked with an obviative marker.
The proximate/obviative system cannot be considered to be a type of case
marking, as the notions of proximate and obviative are not associated with
any semantic role in particular, as is illustrated by the following construc-
tions involving two third-person referents.
(28) Náhe ka’ėškóne é-ohke-véstȧhém-óho

deic.end.dist.an boy 3-eq-help.vta-3>4.3sg.a.4sg.u
néhe ma’háhkėse-ho.
deic.end.prox.an old.man-obv
‘That boy always helps this old man.’
do′ (náhe ka’ėškóne, [help′ (náhe ka’ėškóne, néhe ma’háhkėseho)])

(29) Néhe ma’háhkése é-ohke-véstȧhém-áá’e

deic.end.dist.an old.man 3-eq-help.vta-4>3.4sg.a.3sg.u
náhe ka’ėškóne-ho.
deic.end.prox.an boy-obv
‘That boy always helps this old man.’ / ‘This old man is always helped by
that boy.’
do′ (náhe ka’ėškóneho, [help′ (náhe ka’ėškóneho, néhe ma’háhkése)])
Despite the fact that these sentences include the same arguments and are
semantically equivalent, the arguments show a different proximate/obvia-
tive status. Thus, whereas in (28) the agent is the proximate náhe ka’ėškóne
‘that boy’, and the benefactive is the obviative néhe ma’háhkėseho ‘this old
man’, in (29), there is a shift with respect to the proximate marking. Now,
the agent náhe ka’ėškóneho ‘that boy’ is obviative and the benefactive néhe
ma’háhkése is proximate.
A representation of direct core arguments (28) is provided in Figure 22.3
for a subsequent comparison with that of incorporated arguments in appli-
cative constructions in Figure 22.4 and argument-adjuncts and adjuncts in
Figure 22.5.
As there is no correlation between the proximate/obviative distinction and
semantic roles or syntactic functions, it is necessary to take into account the
information provided by the nominal morphology and the prefix and pro-
nominal suffixes on the verb. Thus, the difference between these two
examples of transitive predication lies in that, while in the direct construc-
tion in (28) the proximate argument náhe ka’ėškóne ‘that boy’ is the actor and
the obviative argument néhe ma’háhkėseho ‘this old man’ is the benefactive,
the opposite occurs in the inverse construction in (29), as the obviative
argument náhe ka’ėškóneho ‘that boy’ is now the actor and the proximate
argument néhe ma’háhkése ‘this old man’ is the benefactive.
SENTENCE
CLAUSE
CORE
RP PRO NUC PRO RP
CORER CORER
NUCR PRED NUCR
N V N
Náhe ka´ škónei éi- ohk - vést hém - óhoi+j néhe ma´háhk sehoj
‘That boy always helps this old man.’

Figure 22.3 Representation of direct core arguments (cf. 28)

SENTENCE
CLAUSE
CORE PERIPHERY
PRO NUC PRO RP CLAUSE
NUCP RPIP CORER
PRED PRED RPPOSS PRO NUCR
P V N N
Nái-htse-vést- homó´hem-ohoi+j vého he-stónahoj tséhmóheeohtséstovetse
‘I will dance with the chief’s daughter at the meeting.’

Figure 22.4 Representation of an applicative construction with a derived two-place verb
SENTENCE
CLAUSE
CORE PERIPHERY
PRO NUC PRO RP RP RP
NUCP CORER CORER CORER
PRED PRED NUCR NUCR NUCR
P V N N N
Nái- ést - o´tse - n tsei+j m xe´ stóon tsej hoo´hénóva éše va
‘I put the books in the bags yesterday.’

Figure 22.5 Representation of a clause including an argument-adjunct and an adjunct
Cheyenne has a number of applicative constructions where the addition of

a relative root in the form of an initial stem increases the syntactic valence
of the verb and, consequently, allows it to take one more direct core
argument.
(30) Nȧ-htse-vést-ȯhomó’hem-óho vého he-stóna-ho

1-fut-with.in-dance.fta-1>4.1sg.a.4sg.a chief 3.poss-daughter-obv
tsé-h-móheeohtsé-stove-tse.
cnj-pst-meet.vai.imper-ptcp
‘I will dance with the chief’s daughter at the meeting.’
do′ (1sg, [dance′ (1sg)]) ^ do′ (vého hestónaho, [dance′ (vého hestónaho)])
While the predicate ‘dance’ requires only one semantic argument,

the complex verb véstȯhomó’hem ‘dance with’ in (30), formed by an adposi-
tional initial vést- ‘with’ and a transitive verb stem ȯhomó’hem ‘dance’,

cross-references two arguments, namely a first-person singular agent and a

third-person singular comitative vého hestónaho ‘the chief’s daughter’,
through the prefix ná- ‘1’ and the pronominal affix -o, which also includes
a reference to the argument vého hestónaho ‘the chief’s daughter’. Thus,
although it does not refer to a participant semantically required by the
predicate ‘dance’, the RP vého hestónaho ‘the chief’s daughter’ is treated as
an argument of the complex verb véstȯhomó’hem ‘dance with’, yielding a
derived two-place verb.
Likewise, unlike argument-adjuncts (see Section 22.2.5), this RP is cross-
referenced on the verb, and, unlike adjuncts, it is not marked as an
oblique.13 The fact that the cross-reference of the RP vého hestónaho ‘the
chief’s daughter’ is subject to the addition of the adpositional initial vést-
‘with’ makes this argument the closest equivalent to an oblique core argu-
ment. The only difference is the fact that the adpositional initial is not
determined by the verb. Rather, it is predicative and, consequently, it
contributes its own predicate to the clause. It is also of note that the
possessor RP in the genitive phrase is represented like an element branching
from the RP node that agrees in person with the prefix designating the
possessor, namely he- ‘his/her’, which behaves as a core argument and is,
consequently, represented as a daughter of the nominal core.
22.2.5 Adjuncts, Argument-Adjuncts and Adpositional Marking

Adjuncts in Cheyenne can appear in very different guises, for example:
verbal premodifiers (e.g. -ohke- ‘usually’ in (31)), nominal phrases (e.g. méó’ne
‘on the road’ in (31)), free particles (e.g. hó’ótova ‘sometimes’ in (31)), or
dependent clauses (e.g. tséhmóheeohtséstovetse ‘at the meeting’ in (30)). They
function as optional modifiers because they are never cross-referenced on
the verb or incorporated with an adpositional initial into the verb. Further-
more, they may receive oblique (e.g. locative (31), temporal (17) or instru-
mental (see example in note 13)) case marking through the nominal suffixes
-va, -e, or -o, when they are realized through nominal phrases.
(31) Hó’ótóva ná-ohke-ameváen-one méó(’o)-n-e.

sometimes (1)-eq-pass.vta-1>3.1pl.a.3sg.u road-epen-loc
‘Sometimes we pass him on the road.’ (Fisher et al. 2006: 11)
[do′ (1pl, Ø)] cause [ingr overtaken′ (3sgM)]
The locative adjunct méóne ‘on the road’ is formed by the noun méó’o ‘road’
plus the oblique nominal suffix -e, which denotes location. This locative
adjunct is neither cross-referenced on the verb nor incorporated into the
verbal complex as a stem-forming element.
Despite their reference to one obligatory argument of a verb, argument-
adjuncts in Cheyenne are not cross-referenced on the verb, which makes
them resemble adjuncts syntactically. Cheyenne also does not distinguish
adjuncts from argument-adjuncts in terms of marking, as the same oblique

suffixes, -va, -e or -o, are added to both adjuncts and argument-adjuncts

regardless of their syntactic function.
(32) Ná-ést-o’tse-nȯtse mȯxe’ėstóonȯtse hoo’hénó-va

1-in.in-put.fti-1>I.1sg.a.Ipl.u books bags-loc
éše-ē-va.
day-epen-temp
‘I put the books in the bags yesterday.’
[do′ (1sg, Ø)] cause [ingr be.in′ (hoo’hénóva, mȯxe’ėstóonȯtse)]
The verb ‘put’ in (32) requires three obligatory arguments – an agent, a

patient, and a locative. However, táho’tsé ‘put’ only has two direct core argu-
ments, the agent and the patient, as can be evidenced from the TI stem and
the verbal suffix -nȯtse, which codes only two arguments – a first-person
singular animate agent and an inanimate plural patient – but does not reflect
the presence of any marker representing the location. The location is
expressed by an RP, hoo’heno ‘bags’, which is marked as oblique by the nominal
suffix -va,14 but is not coded on the verb, which results in its being marked as
a peripheral oblique. However, unlike adjuncts, such as éšeēva ‘yesterday’,
argument-adjuncts appear to be complemented by a locative adpositional
initial particle such as ést- ‘in’ in (32). This is an adpositional nucleus incorpor-
ated into the nucleus of the core functioning as an applicative marker that
licenses the third core argument, the argument-adjunct hoo’hénóva ‘the bags’.
The stem-forming element reflects the presence of a locative argument and it
is this adpositional element, rather than the locative RP, that is obligatory in
the sentence. The locative particle is an important component of the meaning
of the sentence, and its form is not determined by the verb – the particle ést-
‘in’ can be replaced by -táh(o’k)- ‘on’, -áhto’- ‘under’, etc. This incorporated
particle must therefore be considered to be an adpositional nucleus that
functions as an applicative marker.
Finally, this construction also shows that not all three-place predicates are
realized through a ditransitive construction in Cheyenne.
22.2.6 The Linking Algorithm in Cheyenne

The study of the interaction between the various components of grammat-
ical structure generally shows language-specific variation. Bearing this in
mind, for an examination of systematic relationships in the semantic con-
tent of a predicate, the syntactic behaviour it exhibits, and the pragmatic
background that surrounds the act of communication, a bidirectional analy-
sis of the linking system based on example (33) is presented in Figures 22.6
and 22.7, illustrating how the syntax–semantics–pragmatics interface works
in Cheyenne.
(33) Na-né’ame ná-met-aenóvo mȯxe’ėstoo’o.

1.poss.parents 1-give.ditr-3>1>I.3pl.a.1sg.u.Isg.nmr book
‘My parents gave me a book.’

From semantics to syntax:

SENTENCE
SYNTACTIC
INVENTORY
CLAUSE
4
CORE
RP PRO NUC PRO PRO RP
PRED
Na-néámei ná- mét- aenóvo m xe´ stooók

(1j) (3PLi>1SGj>ISGk)
5
LEXICON ARG INVERSE: 3>1>I PSA: ARG ARG

3
1
ACTOR 2 UNDERGOER NMR
do′ (3 [na-néámeiACV], Ø)]CAUSE[BECOME have′ ( [1sg j ACS], I [m xe´ stooókACV ]]
Figure 22.6 Linking from semantics to syntax in Cheyenne

ACS ¼ accessible; ACV ¼ activated
SENTENCE
PARSER
Direction of the Action? - Inverse CLAUSE

PSA = Undergoer
CORE
1
RP PRO NUC PRO PRO RP
PRED
Na-néámei ná- mét- aenóvo m xe stooók

(1j) (3PLi>1SGj>ISGk)
ACTOR UNDERGOER NMR

LEXICON 3 ACTOR UNDERGOER
2 1
[do′(x i,Ø)] CAUSE [BECOME have′ (y j,zk)]
Figure 22.7 Linking from syntax to semantics in Cheyenne

This linking procedure from semantics to syntax starts with the construc-
tion of the semantic representation of the sentence, which is drawn from
the logical structure of the predicate.
As is the case with other languages, the selection of the actor and under-
goer arguments in monotransitive constructions presents no difficulty in
Cheyenne, as the highest-ranking argument in the logical structure is
selected as the actor and the lowest-ranking argument is selected as the
undergoer. Ditransitive constructions indicate that Cheyenne represents
an instance of what Dryer (1986) called a ‘primary object language’, that is
to say a language where the only pattern that occurs with three-argument
predicates corresponds to the marked selection for undergoer.15 Thus,
when a verb has three arguments, the second-highest, rather than the
lowest-ranking argument in the logical structure is selected as the under-
goer macrorole and, consequently, the lowest-ranking argument is the
non-macrorole direct argument. In example (33), therefore, the leftmost
argument in the logical structure of na-né’ame ‘my parents’ is selected as
the actor, and the second leftmost argument – the participant referring to
the speaker – is chosen as the undergoer instead of the expected rightmost
argument mȯxe’ėstoo’o ‘book’, which now becomes the non-macrorole
argument.
The determination of the morphosyntactic coding of the arguments shows
an interesting specific feature of Algonquian languages, as the choice of a
specific argument on the prefix shows the existence of a privileged syntactic
argument (PSA) in these languages. This PSA involves a pragmatically deter-
mined neutralization of semantic roles for syntactic purposes, as the choice
of the prefix is pragmatically determined – it is only determined by the
Person Hierarchy – and can correspond to different semantic roles
depending on the ranking of the participants on the hierarchy (e.g. agent
(4), patient (19), experiencer (1), recipient (33), etc.). In a direct transitive
construction, the PSA corresponds to the actor, and in an inverse transitive
construction the PSA matches the undergoer. Thus, in example (33), bearing
in mind that it is an instance of inverse construction, the PSA is the under-
goer cross-referenced on the prefix ná- ‘1’.
As regards the syntactic representation of the sentence, the syntactic
template for a three-place predicate like mét ‘give’ must contain a core
including three argument positions, which will be filled by the pronominal
arguments cross-referencing the core arguments, namely the prefix ná- ‘1’
and the portmanteau verbal suffix -aenóvo, which cross-references the three
semantic arguments of the verb.
Finally, the linking of arguments to positions in the syntactic template is
also an interesting feature of Cheyenne. The fact that word order in this
language is pragmatically determined allows for greater flexibility when it
comes to selecting the syntactic template of the sentence because of the
positions (e.g. pre-core slot (PrCS) or any extra-core slot (ECS)) to which the
independent RPs are assigned.

The direction of the linking process from syntax to semantics involves the
interpretation of the overt morphosyntactic form of a sentence and the
deduction of the semantic functions of the elements in the sentence deriv-
ing from such an interpretation, as shown in Figure 22.7.
The information provided by the verbal affix -aenóvo reflects the fact that,
in this inverse construction, the verb is ditransitive and has three direct core
arguments – a third-person plural agent, a first-person singular recipient,
and an inanimate singular theme. The prefix also enables us to identify
which constituent acts as the PSA, since, according to the ranking of the
Person Hierarchy, which favours first person over third and inanimate
person, ná- ‘1’ acts as the PSA of the construction. As regards the identifica-
tion of the semantic macroroles of the core arguments, a ditransitive verb
like mét ‘give’ has three possible candidates for only two macroroles. How-
ever, taking into account the accurate grammatical information provided by
the bound markers ná- ‘1’ and -aenóvo, and considering that Cheyenne
invariably shows the marked undergoer selection in ditransitive construc-
tions, we can then select the proximate participant na-né’ame ‘my parents’ as
actor (‘x’) and the first-person participant as undergoer (‘y’), whereas the
third and last core argument, namely the inanimate participant mȯxe’ėstoo’o
‘book’, becomes the non-macrorole argument (‘z’).
The linking of the core arguments to the corresponding slots in the semantic
representation of the sentence may be problematic in Cheyenne due to the fact
that the number of pronominal affixes in the verb does not necessarily corres-
pond to the number of direct core arguments. This is mainly because the
verbal suffix is generally a portmanteau that cross-references all the direct
core arguments of a verb, and the prefix only codes one of these core argu-
ments, which does not allow for a one-to-one correspondence between direct
core arguments and semantic arguments. Thus, it is necessary to bear in mind
what type of stem the verb presents and what kind of grammatical infor-
mation is provided by each of the pronominal affixes in order to gauge its
number of direct core arguments. It would therefore seem logical to assume
that the portmanteau cross-references all the arguments and the prefix cross-
references one of them again, as represented in Figure 22.7.
The linking algorithm shows the important role played by pragmatics in the
interaction between semantics and syntax in Cheyenne and reveals a series of
grammatical properties: the variable word order exhibited by RPs, the prag-
matically determined PSA, the cross-referencing of arguments through a port-
manteau, and even the lack of correlation between the proximate/obviative
status of referents and their semantic function in non-local contexts.
22.3 The Role of Pragmatics in Focus Structure, PSA Selection

and Macrorole Assignment
The lack of a proper case-marking system and the use of a pragmatically

influenced word order, which is not concerned with the identification of

grammatical relations or semantic roles and, even, not completely deter-

mined by the notion of presupposition and assertion, make the process
of the transfer of information between syntax and semantics and vice
versa more difficult than in other languages. Thus, the goal of this
section is to discuss the role of pragmatics in clause structure, PSA
selection and macrorole assignment by discussing the interaction
between word order, the reference-tracking system of obviation, and
the direct/inverse system.
22.3.1 Word Order and Focus Structure

As discussed in Section 22.2.1, word order in Cheyenne encodes pragmatic
factors rather than argument relations. Therefore, the language exhibits a
wide range of variation in word order, which appears to be sensitive to the
notion of newsworthiness. Specifically, the information is distributed in
descending order of importance and unpredictability in the sentence.
Importantly, the fact that word order in Cheyenne is not as strict as in other
languages, English for example, does not necessarily mean that it is com-
pletely free, since it may be constrained by focus structure.
Before explaining the fundamentals of Cheyenne word order from the
perspective of RRG, it is necessary to examine the way in which the arrange-
ment of discourse functions fits into the RRG model of sentence structure.
The arrangement of the pragmatic categories in a specific linear order has
a functional explanation. The leftmost category (see Figure 22.8), which
includes topic particles and adverbials, equates to the pre-detached position
(PrDP) in syntax, whose position is set off from the clause-internal elements
by a pause. The next category, a focused constituent occupying the pre-core
slot (PrCS) within the clause, corresponds to elements introducing discourse
topics expressing a non-contrastive change of discourse topic. Alternatively,
the same position can host an element introducing a contrast, regardless of
whether this element has been previously mentioned in the discourse (in
SENTENCE
CLAUSE
PrDP PoDP
CORE
PrCS RP (RP)
PRO NUC PRO
PRED
V
Topic Shifting/Information/Contrastive Verbal complex Information/Non-shifting Familiar
Part/Adv Topic Focus Topic/Focus Focus Topic Topic
NEWSWORTHY NON-NEWSWORTHY
Figure 22.8 Information structure-based sentence template

which case it is a contrastive topic) or is newly asserted (contrastive focus).16

Unless it is considered newsworthy, the verbal complex follows the focused
element, forming the core of the clause and serving as the point of reference
for the placement of the other clausal elements within the template. Both
new information and easily inferable, already evoked or predictable, infor-
mation, which serves to recover background information, occupy the post-
verbal position. Finally, the rightmost element, called a familiar topic in
Figure 22.8, can also be used for an afterthought, which is included in order
to reiterate its referent by a way of summary or to help to maintain the
reference.17 This is also separated from the previous clause-internal constitu-
ents by a pause, meaning that it is located in the post-detached
position (PoDP).
The availability of two different positions for informational focus can be
explained with reference to the concept of newsworthiness. Indeed, the
positioning of informational focus before or after the verbal complex
appears to be related to the significance given by the speaker to this
information. However, the fact that word order in Cheyenne is governed
by the principle of newsworthiness and that this notion is not entirely
comparable to the pragmatic notions of assertion and presupposition that
underlie the discourse-pragmatic statuses of focus and topic in RRG,
makes it difficult to integrate the information structure of Cheyenne
into the layered structure of the clause. Nevertheless, the tendency
towards verb-initial word order,18 and the identification of the preverbal
position (PrCS) as the most pragmatically marked position within the
clause (it is the position occupied by the elements considered more infor-
mationally prominent in terms of newsworthiness) are suggestive of a
correlation between word order and the different focus types proposed
in RRG.
Cheyenne appears to use a verb-initial order for predicate focus, the
universally unmarked type of focus structure.
(34) Q: É-tóneto’omenehe-Ø(-he) Richard?

3-what.happen.vai-3sg.u(-if ) Richard
‘What happens to Richard?’
A: é-sáa-mé’ov-ó-he-ho neše
3-neg-find.vta-3>4.3sg.a.4sg.u-neg-4sg.u two
mo’éhno’hāme (Richard).
horse.obv Richard
‘Richard/He does not find two horses.’
The pragmatic presupposition in this type of focus structure includes know-

ledge of a certain topic, and the assertion expresses a comment about the
topic. The most common sentence type in Cheyenne in this context is a
comment–topic structure where the predicate mé’ov ‘find’ and the patient
neše mo’éhno’hāme ‘two horses’ constitute the focus (marked with small caps),
and the agent Richard corresponds to the topic. It is important to remember,

SENTENCE
CLAUSE PoDP
CORE RP
PRO NUC PRO RP

PRED
V
Éi-sáa-méev-ó- he – hoi+j neše mo´éhno´hamej (Richard)
Figure 22.9 Unmarked predicate-focus structure in Cheyenne
SENTENCE
CLAUSE
CORE
PRO NUC RP
PRED
V
é-onén šeotse-Ø na-am ho´héhame
Figure 22.10 Unmarked sentence-focus structure in Cheyenne
however, that it is very difficult to find examples of constructions contain-

ing given or familiar topics occupying a postverbal position, so that leaving
the topical referent Richard unspecified appears to be more common, unless
this previously mentioned information deserves to be reiterated in order to
specify or clarify the reference. Figure 22.9 represents the focus structure
of (34).
The actual focus domain (the heavy solid line in Figures 22.9 to 22.11) in
the unmarked type of predicate-focus structure comprises the core verbal
constituent, if the verb is intransitive, and, additionally, the postverbal ECS,
when there is (are) some overt RP argument(s). Furthermore, as expected,
prosodic prominence falls on the focal constituents é-sáaméevóheho and neše
mo’éhno’hāme ‘two horses’ in decreasing order. If the sentence included the

SENTENCE
CLAUSE
PrCS CORE
RP PRO NUC
PRED
V
Na-amaho´héhame é-onén še-otse-Ø
Na-am ho´héhame
Figure 22.11 Unmarked narrow-focus structure in Cheyenne
familiar topic Richard, it would be represented as a clause-external topic at

the PoDP. In that case, this topical constituent would be out of the potential
focus domain (the broken line in Figures 22.9 to 22.11) and those that
follow), which is coextensive with the clause in Cheyenne.
While it is not so common, the positioning of a topic such as Richard in
the pragmatically marked preverbal position – where it would be activated
as the discourse topic – would illustrate an instance of a marked SV(O)
word order for this focus type. It is of note that, due to its newsworthy
nature, this constituent would be given greater prominence than the
verbal complex.
The second type of focus structure is sentence focus, where no pragmatic
presupposition is evoked, and all information is new and, therefore, in focus.
(35) Q: É-ta-tónėsóotse-Ø éše-ē-va?

I-trans-what.happen.vii-Isg.u day-epen-temp
‘What happened yesterday?’
A: é-onénėše-otse-Ø na-amȧho’héhame.
I-broken.in-become.fii-Isg.u 1.poss-car
The most common word order pattern in this situation, which places the
verb in clause-initial position and the RP in postverbal position, represents
the unmarked structure of this focus type because there is no element
occupying the PrCS slot. Figure 22.10 represents the focus structure of (35).
As there is no presupposition in this structure, the assertion extends over
the entire proposition, as is indicated by the actual focus domain in
Figure 22.10. Both the core and the subject RP in the ECS are associated
with special intonational realization, although its intensity decreases as it
approaches the clause-final position.

Placing the argument RP na-amȧho’héhame ‘my car’ in the pragmatically

marked preverbal position would also be acceptable in this focus structure,
but the word order would represent a case of marked sentence focus. In this
case, the argument RP na-amȧho’héhame ‘my car’ would be highlighted by the
speaker, possibly due to the fact that it represents the new discourse topic.19
The difference between predicate-focus and sentence-focus structures is
therefore both syntactic and prosodic. Unlike sentence focus sentences,
predicate focus sentences (34) have at least one topical element, which is
preferably only cross-referenced by the pronominal affixes on the verb, but
can also occur in the PoDP slot when overtly realized.20 In the latter case,
this topical element is not stressed, unlike the postverbal RP in the sentence
focus construction (35).
The third type of focus structure is narrow focus, where the focus domain
is limited to a single constituent.
(36) Q: É-onénėše-otsé-nėse ne-no’ka’éesó’hestȯtse.

I-broken.in-become.fai-Isg.u.infr 2.poss-motorbike
‘I heard / It is said that your motorbike broke down.’
A: na-amȧho’héhame é-onénėše-otse-Ø.
1.poss-car I-broken.in-become.fii-Isg.u
The narrow-focus construction generally includes an element in the PrCS,

for example na-amȧho’héhame ‘my car’ in (36). Unlike English, Cheyenne
tends to mark narrow-focus constituents both syntactically and prosodi-
cally, with the narrow-focus constituent generally occurring in preverbal
position – as the information it provides is normally considered by the
speaker to be essential to the hearer – where it receives special prosodic
prominence. Figure 22.11 represents the focus structure of (36).
The constituent under narrow focus is the only one that is prosodically
stressed; hence, the actual focus domain does not generally extend into the
core in Cheyenne. Thus, in Cheyenne, the unmarked narrow-focus position
for single constituents with focal properties is the PrCS. However, when
such focal material occurs in other positions, the marked narrow-focus
structure is called for. This situation is rather infrequent, as the narrow-
focus constituent provides the most important information and, conse-
quently, the information it provides is generally considered newsworthy,
but it might still occur in constructions including a transitive verb and two
overt argument RPs where either the verb or either of the RPs could be
considered by the speaker to convey more newsworthy information. In that
case, the actual focus domain would move elsewhere within the clause.
Thus, narrow focus on a newsworthy element is a case of unmarked
narrow focus, while narrow focus on a non-newsworthy element is a case
of marked narrow focus.
To summarize, while in Cheyenne the influence of the discourse-
pragmatic concept of newsworthiness on the syntactic arrangement of

constituents within a clause is paramount, the word order is not free, as it

also appears to encode discourse-pragmatic functions such as focus and
topic. Cheyenne selects verb-initial order as its unmarked word order in
the predicate- and sentence-focus structures. The default interpretation of
elements in the PrCS in these constructions is that of marked focus, for
example contrastive narrow focus (but note that contrastive topics can also
occur there, see Shimojo 2009, 2010, 2011 and Chapter 11 on information
structure). Finally, prosodic prominence is always associated with the
placement of focal elements in the PrCS or, in the absence of an RP
occupying such a slot, in the core-initial position.21 Thus, assuming that
in predicate- and sentence-focus structures the unmarked focus position
for RPs in Cheyenne is postverbal and the marked focus position for RPs is
preverbal – more specifically in the PrCS – the potential focus domain
extends over the entire clause in all types of focus structure, but the actual
focus domain is: (1) from the core to the end of the clause in unmarked
predicate- and sentence-focus structures; or (2) just the PrCS in the
unmarked narrow-focus type.
Finally, while the notion of newsworthiness fails to correspond to the
dichotomies topic vs. focus or even given vs. new information (Leman
1999), a correlation between newsworthy information and focus emerges
from the analysis, as evidenced by the fact that, excluding the variable
positioning of informational focus, the preferred constituent order in Chey-
enne is one where focal material (informationally prominent or contrastive)
precedes the verbal complex, and the topic material (known or presupposed)
appears after the verbal complex.22
22.3.2 The Proximate/Obviative System

Now that the link between word order and information packaging in Chey-
enne has been established, this section turns to an analysis of the proximate/
obviative system. According to Aissen (1997: 709), this marking serves as a
reference-tracking system whereby the language is able to differentiate mul-
tiple non-local participants in such a way that, in a discourse span involving
two or more third-person animate arguments,23 only one of them is
unmarked and placed in the foreground (the proximate) and all others are
morphologically marked and relegated to the background (the obviative).
When a discourse span begins, the most pragmatically salient participant,
the proximate, is usually promoted to a preverbal position, and the obvia-
tive, which refers to the other participants, tends to occupy a postverbal
position. Taking into account the fact that the preverbal position is gener-
ally reserved for newsworthy information and receives special prosodic
prominence in Cheyenne, it might be logical to assume a certain correlation
between the notions of proximate and focus, on the one hand, and those of
obviative and topic, on the other. This correspondence is not absolute,

however, as it is possible to place an obviative argument in clause-initial

position even when this position has inherently focal properties, as is the
case with an obviative interrogative pronoun, for example. As discussed in
Section 22.2.4, there is no correlation either between the distinction
between proximate and obviative marking, which differentiates two non-
local participants in terms of discourse salience, and semantic roles or
syntactic functions (cf. 28–29).
While it is not completely clear what factor determines the choice
between proximate and obviative, discourse salience must be one such
factor, in the sense that the proximate is the most prominent participant
of the text or the topic of discourse. The first thing a speaker must do before
formulating a message is decide which of the third-person referents is the
central participant in the text. Subsequently, they will leave this proximate
argument, which tends to appear in preverbal position, unmarked, and
mark all other third-person animate participants, which generally occupy
postverbal positions, as obviative.
Indeed, it is very common to find long stretches of narrative text where
the same referent, considered by the speaker as the most important partici-
pant, is left unmarked as proximate.
(37) 1. Kȧsovááhe é-h-ne’-éva-ame-vé’otsė-hoo’o.

young.man 3-pst-cisl-back-continue-go.on.warpath.vai-3sg.a.pret
‘A young man returned from scouting.
2. É-š-kȧhane-otsé-hoo’o.
3-pst-tired.in.become.fai-3sg.u.pret
He was tired.
3. É-x-ho’ėhahtse-Ø tósa’e nėhéóhe o’hé’e.
3-pst-make.fire.vti-3>I.3sg.a.Isg.u somewhere there river.loc
He made a fire somewhere there at a river.
4. Tsé-’-ēe-hoo’ė-se nėhéóhe tóxė(ha)-ho’ėsta-va
cnj-pst-around-stay.vai-3sg.u.ind there beside-river.loc
As he was sitting there beside the fire,
É-x-ho’ėhó’t-aehoono nevá’ėsesto.
3-pst-come.to.vta-4>3.4sg.a.3sg.u.pret someone.obv
someone came up to him.
5. “Kȧsovááhe, né-ta-néhovan-ȧhtse-ma”,
young.man 2-let-wrestle.vta-recp-12
é-x-het-aehoono.
3-pst-say.vta-4>3.4sg.a.3sg.u.pret
“Young man, let’s wrestle!” he said to him.
6. É-h-néhovan-ahtsé-sesto.
3-pst-wrestle.vta-recp-3pl.rep
They are said to have wrestled.
7. É-s-tsėheta’é(e’tov)’oe-sesto tsé-’-a’enó’ne-otse-Ø
3-pst-face.vta-4>3.4sg.a.3sg.u.rep cnj-pst-dark.in.become.fii-Isg.a.ind
He is said to have been pushing him towards the dark.

8. Hápó’e, é-s-tsėheta’é(e)’(t)ov-ósesto tsé-h-vōho’ėho’ta-tse.

likewise 3-pst-face.vta-3>4.3sg.a.4sg.u.rep cnj-pst-light.vii-Isg.ind
Likewise, he is said to have pushed him toward the light.
9. É-’- anȧ’ham-osesto.
3-pst-throw.down.vta-3>4.3sg.a.4sg.u.rep
He is said to have thrown him down.
10. “Né-hō’tah-e”, é-x-het-aesesto [. . .].
2-beat.vta-2>1.2sg.a.1sg.u 3-pst-say.vta-4>3.4sg.a.3sg.u.rep
“You beat me”, he is said to have said to him [. . .].’
(Leman 1980a: 37)
The fact that the proximate/obviative distinction serves to track topic

continuity across a discourse span can be observed in (37), an extract
from an oral narrative. The main discourse topic, kȧsovááhe ‘a young
man’, is established from the very outset, as it is placed in preverbal
position, a pragmatically important position in Cheyenne. Once a refer-
ent has been chosen as the main character in the story, it becomes the
proximate argument, and all other third-person participants receive
obviative marking – which can be observed in the nominal suffix (e.g.
nevá’ėsesto [obv] ‘someone’) and in the verbal suffix, (e.g. é-xhetaehoono ‘he
[obv] said to him [prox]’). Thus, for example, the argument nevá’ėsesto
‘someone’ (s. 4) is introduced as new information in sentence 4, but, very
possibly owing to its indefinite character, is not considered by the speaker
to have pragmatic salience. Thus, it is marked as obviative and placed in
postverbal position (note that the proximate counterpart would be
nevá’ėsėstse [prox] ‘someone’). This implies that there is no shift in the
discourse-pragmatic status of these two referents and, consequently, that
the proximate status of the argument kȧsovááhe ‘a young man’ remains
unaltered throughout the stretch of discourse under examination (the
obviative form would be kȧsováaheho [obv] ‘a young man’), as can be
observed in sentences 2 and 3, and in the dependent clause in sentence 4.
However, it is also possible to find texts where the proximate referent
changes very frequently, even from clause to clause.
(38) 1. Vé’ho’e naa xaa-vo’ėstane é-’-ēe-néše-ohtsé-sesto [. . .].

white.man and Indian 3-pst-around-continue-go.vai-3pl.a.rep
‘A white man and an Indian were going together.
2. [. . .] Naa tsé’tóhe xaa-vo’ėstane
and deic.prox.exo.an Indian
é-’-eše-aahtse’-tótoéše-na-sėtse.
3-pst-lie-already-lie.with.eyes.open.vai-?-3sg.u.rep
[. . .] this Indian is said to have been already lying with his eyes open.
3. Naa vé’hoé é-’-osee-hóhta’ȧhané-tano-sėstse.
and white.man 3-pst-very-tell.story.vai-want.fai-3sg.a.rep
And the white man is said to have really wanted to tell his story.

4. É-h-nė-hetó-sesto tsé’tóhe xaa-vo’ėstan(e)-óho,

3-pst-ana-tell.vta-3>4.3sg.a.4sg.u deic.prox.exo.an Indian-obv
He told this Indian:
5. “ná-ta-ovóe-hósésta na-ováxestȯtse!”
1-let-first-tell.vti.3>I.3sg.a.Isg.u 1.poss-dream
é-x-het-ósesto.
3-pst-tell.vta-3>4.3sg.4.sg.u.rep
“Let me first tell my dream!” he is said to have told him.
6. Naa tsé’tóhe xaa-vo’ėstane
and deic.prox.exo.an Indian
é-h-pėhév-átsėstá-nȯse.
3-pst-good.in-regard.fti-3>I.3sg.a.Isg.u
and this Indian thought well of that.
7. Naa néhe vé’ho’e é-’-asėst-(h)óhta’hane-sėstse.
and deic.end.prox.an white.man 3-pst-start-tell.story.vai-3sg.a.rep
And that white man started to tell his story.
8. [. . .] Naa nėhéóhe tsé-s-ta-éšė-ho’óhta’ȧhanė-se,
and there cnj-pst-trns-already-arrive.in.storytelling.vai-3sg.a.ind
[. . .] And when he got to that point in the story,
tse’tóhe xaa-vo’ėstane é-h-nė-het-ósesto
this Indian 3-pst-ana-tell.vta-3>4.3sg.a.4sg.u.rep
this Indian said to
tsé’tóhe vé’hó’e, [. . .].’
deic.end.prox.an white.man.obv
this white man [. . .].’
(Leman 1980a: 40)
In this different story, shifts of proximate/obviative status occur very

frequently because the speaker selects two discourse topics, rather than
only one as in the previous story. The speaker considers the two referents,
vé’ho’e ‘white man’ and xaa-vo’ėstane ‘Indian’, as equally important. This is
supported by the first sentence of the story, where both referents appear in
preverbal position forming a complex RP. From this moment onwards,
whenever the speaker refers separately to either of them in the story, the
referent being referred to receives the proximate status and occupies the
preverbal position, for example tsé’tóhe xaa-vo’ėstane ‘this Indian’ in sentence
2 and vé’hoé ‘the white man’ in sentence 3. This leads to a rapid series of
changes of discourse topic and of the discourse-status of the arguments, so
that xaa-vo’ėstane ‘Indian’ is the proximate argument in sentences 2, 6 and 8,
whereas vé’ho’e ‘white man’ is the proximate argument in sentences 3 and 7.
Each of these two referents also receives obviative marking in one sentence,
that is to say xaa-vo’ėstan(e)-óho [obv] ‘Indian’ in sentence 4 and vé’ho’e [obv]
‘white man’ in sentence 8. The fact that they are not considered pragmatic-
ally important in these two sentences is supported by their obviative
marking – reflected both in the nominal suffixes and the verbal suffixes –
and their placement in the postverbal position.

The analysis of the proximate/obviative distinction in these two texts

shows that the main topic of discourse in Cheyenne tends to be proximate,
and that the proximate/obviative status of referents remains unaltered until
a new discourse topic (most commonly, a new focal element or an old
topical element that is reintroduced into the discourse) is established,
generally bringing about a shift in the discourse span and leading to a
new proximate/obviative assignment.
However, as the proximate and obviative distinction is not invariably tied
to macrorole assignment, it cannot by itself help us to determine which
macrorole corresponds to each argument. This can be observed, for example,
in sentences like 5, 7, 8, 9 and 10 in (37) or 5 in (38), where verbs cross-
reference two third-person animate arguments with non-overt RPs, whose
interpretation may confuse the hearer.
22.3.3 The Direct/Inverse System

As seen in Section 22.2.1, the Person Hierarchy and the Semantic Function
Hierarchy work together as a single unified system in order to establish the
link between syntactic and semantic information, as shown in Figure 22.12
(see also Wolvengrey 2011: 57–63).
The interaction between the two hierarchies establishes a correspondence
between actors and local participants, on the one hand, and undergoers and
non-local participants, on the other, in such a way that a harmonic align-
ment between the two hierarchies leads to a direct construction, and a
disharmonic alignment between the two scales results in an inverse
construction.
Both the Person Hierarchy and the Semantic Function Hierarchy rank
their elements in terms of their inherent pragmatic salience or prominence.
However, each hierarchy appears to be based on a different notion of
salience or prominence. While the Semantic Function hierarchy is sensitive
to the degree of animacy of the participants,24 the Person Hierarchy may be
governed by several pragmatic factors, such as animacy or topicality, as it
ranks local over non-local participants and animate arguments over inani-
mate arguments.25
As discussed in Section 22.2.1, the person represented by the prefix in an
intransitive sentence always corresponds to the only direct core argument of
the verb, so that the suffixal morphology is only concerned with the
Speech Act forms non-Speech Act forms

Person: 2 > 1 > X > 3 > 4 > I
(Semantic Macrorole: Actor > Undergoer)
Semantic Function: Agent > Recipient / Benefactive > Patient / Theme
higher salience ---------------------------- lower salience
Figure 22.12 Interaction between the Person and Semantic Function hierarchies

distinction between singular and plural number (see examples 11 and 13) or
obviative marking (see 15). However, in transitive constructions, where more
than one participant is involved, the prefix only cross-references the most
pragmatically salient argument in terms of the Person Hierarchy.
(39) Né-vóom-ȧtse.
2-see.vta-1>2.1.sg.a.2sg.u
‘I saw you.’
(40) Né-vóom-e.
2-see.vta-2>1.2sg.a.1sg.u
‘You saw me.’
In these examples the prefix né- ‘2’ is selected as the prefix heading the verbal
complex, thereby acquiring a special syntactic status, namely the PSA of the
construction. Neither the prefix né- ‘2’ nor the pronominal suffix -ȧtse or -e
can determine the semantic roles of the arguments of the verb by them-
selves. The prefix cross-references only the higher-ranking person, but is not
indicative of semantic role on its own – it cross-references the undergoer in
(39) and the actor in (40) – and the portmanteau formed by the theme
marker plus the pronominal affixes tends to provide information about
the direction of the construction, and to cross-reference the direct core
arguments of the verb, but is sometimes ambiguous, especially in the
Conjunct order. An examination of both elements, therefore, becomes neces-
sary in order to link arguments and semantic roles. Thus, the prefix in both
examples is a second-person participant; the portmanteaus -ȧtse and -e indi-
cate that (39) and (40) are an inverse and a direct construction respectively,
and that these constructions include a first-person singular actor and a first-
person singular undergoer, respectively.
The direct/inverse mechanism is especially useful when the point of view
changes for pragmatic reasons, and the core arguments are not lexicalized
as RPs.
(41) Ná-mó’ot-óneo’o Heévȧhetaneo’o

1-invite.vta-1>3.1pl.a.3pl.u Oklahoma.Cheyennes
‘We invited Oklahoma Cheyennes.’
naa ná-vés(e)-e’hanám-aeneo’o.
and 1-with.in-eat.fta-3>1.3pl.a.1pl.u
‘. . . and they ate with us.’ (Leman 1980a: 71)
In this example, the pronominal affixes -ae-ne-o’o, which provide unambigu-

ous grammatical information about each of the core arguments of the verb,
help us to identify that the clause ná-vés(e)e’hanámaeneo’o is an inverse con-
struction, as it involves a third-person plural animate participant acting on a
first-person plural animate participant.
To summarize, as the morphology of Cheyenne has no case distinctions,
the only way to distinguish actor from undergoer in a transitive construc-
tion – see examples 5, 7, 8, 9 and 10 in (37) or 5 in (38) – is through the joint

analysis of the direct/inverse system, the Person and Semantic hierarchies

and the proximate/obviative status. The direct/inverse system provides a
successful link between the arguments of a predicate and their semantic
roles by encoding the interaction between these hierarchies. Thus, this
efficient direct/inverse system, supported by the proximate/obviative distinc-
tion in non-local environments and without recourse to a fixed syntactically
oriented rigid word order, plays a similar role to case marking in other
languages in order to link semantic arguments with specific semantic
functions in Cheyenne.
Finally, once macroroles have been assigned to semantic arguments, the
speaker must select an appropriate word-order pattern following the
principle of newsworthiness. In accordance with this pragmatic principle,
the preverbal position in Cheyenne clauses is considered newsworthy, as it is
here that discourse-prominent material is placed, and, by contrast, less
discourse-salient information is placed in postverbal position.
22.4 Conclusion
The goal of this grammatical sketch has been to explore sentence structure
in Cheyenne from the perspective of RRG through the analysis of a range of
grammatical issues serving both to reveal the core components shared by all
languages and to highlight those that are specific to Cheyenne grammar.
The analysis of Cheyenne grammar in Section 22.2 showed that there is no
evidence for the postulation of grammatical relations in addition to seman-
tic predicate–argument relations, save for a pragmatically influenced PSA
that is represented by the verbal prefix. It also indicates the hierarchical
scope order of operators in Cheyenne, which is broadly defined in accord-
ance with the RRG proposal. Finally, it reveals the fundamental role played
by pragmatics, especially in terms of argument coding, macrorole assign-
ment and word order.
Section 22.3 argued that, despite the word-order variability displayed by
Cheyenne, it is possible to integrate information structure into clause struc-
ture and explore the intricate mechanism used by this language to accom-
modate semantic information into syntactic structure. Cheyenne links
arguments and particular semantic roles through an intricate mechanism
consisting of two components, which work in combination with the prox-
imate/obviative distinction in contexts involving non-local participants.
These components are illustrated by: (1) the correlation between the Person
Hierarchy and a Semantic Function Hierarchy and (2) a binary system of
verbal direction expressed by a theme marker or direction-marking mor-
pheme on the verb, which is complemented by the grammatical informa-
tion about the participants provided by the pronominal affixes. The direct/
inverse system works along with the reference-tracking system of obviation,

interrelates the Person and Semantic Function Hierarchies, and acts as a

mediator between all grammatical components. It is thus responsible for
the interplay between syntax, semantics and pragmatics. The efficiency
shown by this system in coding the morphosyntactic properties of argu-
ments from the semantic information and in linking morphosyntactic
coding to macrorole assignment accounts for the existence of the ‘relatively
free’ word order in Cheyenne and the absence of a true case-marking system.
All in all, these findings support the assumption of RRG that syntax is
underpinned by semantic and pragmatic factors. Furthermore, the analysis
of Cheyenne grammar highlights the validity of RRG as a framework for the
analysis of highly polysynthetic languages.
References
Aissen, Judith. 1997. On the syntax of obviation. Language 73(4): 705–750.

Bloomfield, Leonard. 1946. Algonquian. In C. Osgood and H. Hoijer (eds.),
Linguistic structures of Native America (Viking Fund Publications in Anthro-
pology 6), 85–129. New York: Viking Fund.
Corral Esteban, Avelino. 2014. An analysis of transitivity in Cheyenne. Acta
Linguistica Hungarica 61(4): 379–422.
Corral Esteban, Avelino. 2017. Multi-verb constructions in Cheyenne. In B.
Nolan and E. Diedrichsen (eds.), Argument Realization in Complex Predicates
and Complex Events: Verb-Verb Constructions at the Syntax–Semantics Interface
(Studies in Language Comparison Series), 305–346. Amsterdam: John
Benjamins.
Dryer, Matthew S. 1986. Primary object, secondary objects, and antidative.
Language 62: 808–845.
Firbas, Jan. 1992. Functional Sentence Perspective in Written and Spoken Communi-
cation. Cambridge: Cambridge University Press.
Fisher, Louise, Wayne Leman, Leroy Pine Sr. and Marie Sanchez. 2006.
Cheyenne Dictionary. Lame Deer, MT: Chief Dull Knife College.
Hockett, Charles F. 1996. Voice in Algonquian verbs: A suggestion. In John D.
Nichols and Arden C. Ogg (eds.), nikotwāsik iskwāhtēm, pāskihtēpayih! Studies
in Honour of Hans Christoph Wolfart. Winnipeg: Algonquian and Iroquoian
Linguistics: 257–261.
Leman, Elena. 1999. Word Order of Major Constituents in Cheyenne Narratives.
Dallas, TX: Summer Institute of Linguistics.
Leman, Wayne. 1980a. Cheyenne Texts: An Introduction to Cheyenne Literature
(Occasional Publications in Anthropology, Linguistics Series No. 6). Gree-
ley, CO: Museum of Anthropology, University of Northern Colorado.
Leman, Wayne. 1980b. A Reference Grammar of the Cheyenne Language, Vols. 1
and 2 (Occasional Publications in Anthropology, Linguistics Series No. 5).
Greeley, CO: Museum of Anthropology, University of Northern Colorado.

Leman, Wayne. 1987. Náévâhóó’ôhtséme / We are going back home: Cheyenne

History and Stories Told by James Shoulderblade and Others. Memoir 4. Winni-
peg: Algonquian and Iroquoian Linguistics.
Meeussen, Achilles E. 1962. The independent order in Cheyenne. Orbis 11:
260–288.
Mithun, Marianne. 1987. Is basic word order universal? In Russell S. Tomlin (ed.),
Coherence and Grounding in Discourse, 281–328. Amsterdam: John Benjamins.
Petter, Rodolphe. 1952. Cheyenne Grammar. Newton, KS: Mennonite Publica-
tion Office.
Russell, Dale W. 1987. Cheyenne Verb Agreement in GPSG. Doctoral dissertation,
University of Illinois at Urbana-Champaign.
Shimojo, Mitsuaki. 2009. Focus structure and beyond: Discourse-pragmatics
in Role and Reference Grammar. In Lilián Guerrero, Sergio Ibáñez Cerda
and Valeria A. Belloro (eds.), Studies in Role and Reference Grammar, 111–141.
Mexico: Universidad Nacional Autónoma de México.
In Wataru Nakamura (ed.), Proceedings of the 10th International Conference on
Role and Reference Grammar (RRG 2009), 315–335.
In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar,
Siewierska, Anna (ed.). 1997. Constituent Order in the Languages of Europe. Berlin:
Mouton de Gruyter.
Pragmatics Interface. (Studies in Language Companion Series 105). Amster-
Van Valin, Robert D. Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning
Wolvengrey, Arok. 2011. Semantic and Pragmatic Functions in Plains Cree Syntax.
MA thesis, University of Amsterdam.
Notes
1 The prefix selecting the most pragmatically salient person only occurs in the
Independent order (except in the conjectural mode, which is introduced by
the prefix mó-). The prefix in the Conjunct order conveys other grammatical
meanings (e.g. tsé- ¼ realis, vé’- ¼ irrealis, mȧh- ¼ potential, hó’- ¼ iterative,
momóxe- ¼ optative, etc.). The independent and the conjunct verbal orders
generally occur in main and subordinate clauses, respectively.
2 There are two main types of directional particle: (1) the cislocative neh- and
translocative ta-, which serve to express the locational or temporal deixis

of the action, with the speaker or the speech act itself as the deictic locus;
and (2) phoric particles, such as the anaphoric and cataphoric particles nė-
and tsė-, which serve to link elements in discourse, namely a place old or a
place new in discourse respectively, and appear close to the prefix. There
are also other directional initial particles such as he’am- ‘up’, e’e- ‘up’, an
(ȯhe)- ‘down’, ést- ‘into’, sé’- ‘into’, hóe- ‘out’, etc., which always occupy the
rightmost position preceding the verb.
3 Negation in Cheyenne is commonly expressed through two separate and
complementary particles, namely sáa-, a preverbal particle, and -hé, a
postverbal particle that normally occurs between the theme marker and
the pronominal affixes.
4 For the sake of simplicity and because it is not always possible to separate
the theme marker from the pronominal affixes on a verb, both parts will
be analysed as a portmanteau suffix throughout the chapter. In order to
indicate the direction of constructions including transitive verbs, a nota-
tion such as ‘1>3’ will be used, for example to indicate a first-person actor
and a third-person undergoer. For intransitive verbs, its only argument
will be referred to as ‘1’, for example to indicate that the actor or under-
goer is the first person.
5 Adjectives in Cheyenne are always realized as bound particles that
attach to verbs (see example 38(6), in section 22.3.2) or nouns (11) and
behave like adverbs or adjectives in English, respectively. Likewise, they
can also form non-verbal predicates when they are attached as initials
to final verb stems in attributive constructions (see 13). By contrast,
pronouns (see 37(4)), numerals (see 34), adpositions (see 37(4)), and
some adverbs (see 37(3)) are realized as free particles in Cheyenne
and, consequently, excluded from the possibility of functioning
as predicates.
6 Albeit less frequently, other types of identificational sentence can be
expressed in Cheyenne by means of a linking or copular verb (a), the
combination of a preverbal particle and a nominal predicate (b), or even
a verb-less clause (c).
E.g. a. Maah(e)-ótse éhó’tánėstse na-mȧheó-n-e.

arrow.pl I-be.here.vii-Ipl.u 1.poss-house-epen-loc
‘The arrows are in my house.’
be.in′ (na-mȧhēō’o, maah(e)ótse)
b. John ná-he-vésenéhe(‘tov)-nȯtse.
John 1-have-friend.vta-1>3.1sg.a.3sg.u
have.as.friend′ (1sg, John)
‘John is my friend.’ (lit. ‘I have John as friend.’)
consider′ (1sg, [be′(John, [na-(ve)séné])
c. Hé’tóhe máto na’ėstse hóhta’heo’o.
deic.exo-prox-inan also one story
‘This is also another story.’ (Leman 1987: 314)
be′ (hé’tóhe, [na’ėstse hóhta’heo’o])

7 Each verb type has morphological paradigms showing different theme

markers and pronominal affixes. This information is available in a
number of sources (Petter 1952; Meeussen 1962; Leman 1980b; Russell
1987; Corral Esteban 2014).
8 Other ditransitive constructions in Cheyenne consist of a verb plus a
final verb stem, such as, for example: -’seh, in causative constructions,
-omótah, in benefactive constructions, etc.
9 See Section 22.2.6 for more detailed information on undergoer selection
in Cheyenne ditransitive constructions.
10 An aspect of ‘weather’ verbs that is difficult to explain is the fact that,
unlike the impersonal construction, they do not seem to reflect M-
transitivity in the form of their verb stem, as they do not include any
morphological marker indicating that the syntactic valence has been
reduced. A possible solution would be to assume that the consideration
of the single core argument of ‘weather’ verbs is culture-specific, and
that, for example, some cultures understand that there is a non-specific
agent carrying out the activity denoted by the predicate.
11 Applicative constructions (30) would be an exception to this assumption,
where the addition of a relative root, in the form of an initial, increases
the syntactic transitivity and the number of macroroles without altering
the argument structure.
12 This construction appears to resemble the English-style passive in the
sense that it shows the promotion of the patient and the demotion of the
agent, as reflected in the common preverbal position of the former and
its cross-reference by the prefix, on the one hand, and the omission of the
latter, on the other. However, if we assume that the theme markers -Ø
and -án indicate direct and inverse constructions, then direct forms
would not be fully passive, since the patient would not be promoted to
subject status, even though the actor is obligatorily demoted.
13 Adjuncts can also occur in constructions including a relative root, with
the fundamental difference that they are not cross-referenced on the
verb:
E.g. Mȧhaemenȯtse náve’šėpénȯhanȯtse ho’honáéva.

corn.pl 1-by.means.of-grind.vti-1>I.1sg.a.Ipl.u rock.ins
‘I ground corn with a rock.’
The AI verb penȯha ‘grind’ only codes a first-person singular agent and an
inanimate plural patient.
14 It is very difficult to come to a decision about the status of the suffix -va.
On the one hand, given there is an applicative (incorporated adposition)
licensing hoo’hénóva ‘in the bags’ in (32), it does not seem likely that this
applied argument may be marked by a postposition. On the other hand,
the fact that the suffix -va may occur on a variety of argument-adjuncts
and adjuncts with different meanings (e.g. time, location, instrument)

appears to suggest that it is not a case marker in the usual sense. In view
of the above, it seems plausible to consider it a kind of generalized
oblique marker, with no inherent semantics of its own.
15 The evidence for the marked undergoer choice in ditransitive construc-
tions comes from both coding (indexing and ordering of bound forms)
and behavioural properties (reflexivization, reciprocalization, and direct/
inverse marking) (Corral Esteban 2014: 403–415). On the one hand, the
ditransitive verbal paradigms resemble their TA more than their TI
counterparts, and the pronominal markers cross-referencing the recipi-
ent occur closer to the verb than those cross-referencing the theme. On
the other hand, the reflexive/reciprocal suffix only permits coreference
between actor and recipient, but not between actor and theme, and
direct/inverse marking shows a clear secundative bias since the impossi-
bility of constructing ditransitive sentences with inanimate recipients
that are prototypically human, or first- or second-person themes that are
prototypically non-human, implies that only the person of either the
actor or recipient, but not that of theme, can appear as a prefix.
16 I am assuming that focal elements occupy the PrCS rather than the PrDP
in Cheyenne because there is no intonation break between most of the
preverbal RPs and the verbal complex, which suggests that they are part
of the clause.
17 Albeit infrequently, it is also possible to find left-dislocated familiar
topics in Cheyenne.
18 This assumption concerning word order is based on my analysis of word
order patterns in independent clauses found in thirty-five Cheyenne
stories told by different native speakers, which were randomly taken
from Leman (1980a). The analysis revealed that the most common sen-
tence type involves no explicit RPs, and the second most common type
involves one RP. The analysis of intransitive clauses including an RP
shows that the most common word-order pattern is SV (63.48%), which
appears to be more related to newsworthiness (new discourse topic:
83.01% vs. continuing topics: 16.99%) than to givenness (new informa-
tion: 59.47% vs. old information: 40.53%). In transitive clauses including
two overt RPs, the most common word-order pattern is SVO (86.67%),
followed far behind by OVS (8.89%). In the more common situation
displayed by transitive clauses, that is involving an object RP, the most
common pattern is VO (66.41%)).
19 This situation also appears to hold in presentational sentences, where all
the information is asserted and nothing is taken as presupposed. While
Cheyenne appears to follow a verb-initial pattern as its basic, but not
obligatory, word order, sentences that include a direct core argument in
preverbal position also appear to be very common at the beginning of a
story or conversation, because of the speaker’s interest in establishing a
discourse topic from the outset.

20 As discussed above, albeit rather infrequently, this topical element could

also occur in the PrCS if it is considered newsworthy.
21 Except when a single constituent is foregrounded by the speaker, the
verb will be the most pragmatically salient element, as is usual in a
sentence-focus construction.
22 This word order appears to disprove the general assumption that, in
terms of information structure, it is cross-linguistically more common
for topical participants to precede focal participants (Firbas 1992; Sie-
wierska 1997; among others).
23 Owing to language attrition, save for some pronouns, it is not possible to
differentiate proximate and obviative inanimate arguments in Cheyenne
nowadays.
24 The higher-ranking semantic roles (agent, recipient or benefactive) are
linked with animate participants, while the lowest-ranked semantic role
(patient) is associated with inanimate participants.
25 Additionally, the specific preference of second person over first person
may also be motivated by cultural factors (e.g. the tuistical, rather than
ego-focused, nature of the Algonquian culture, politeness conventions,
empathy or modesty, among others).

23
of Yimas (Lower Sepik,
Papua New Guinea)
William A. Foley
Abbreviations

ing-rules.php) with the following additions and modifications:
ASP aspect PAST past tense

CLM clause linkage marker PC paucal
DEIC deictic PERF perfective
DIR directional PN proper noun
DL dual PRES present tense
ELEV elevational POT potential mode
FR.DIST far distal PRO bound pronoun
HAB habitual RED ?reduplicative?
INACT inactivated RM remote
MOD modality RX referential expression
N noun SEQ sequence
NINCORP incorporated noun TNS tense
NFN non-finite V verb
NR.DIST near distal VAL valence
NUC nucleus I-X noun class
NUM number
23.1 Yimas as a Polysynthetic Language
Yimas is a morphologically highly complex polysynthetic language spoken

in the Sepik basin region of the northern swampy lowlands of Papua New
Guinea and is one of six languages in the Lower Sepik sub-family of the
Lower Sepik-Ramu family (see Foley (1986, 2005, 2017a) for comparative
studies of this family). It has a typical profile for a polysynthetic language:

862 WILLIAM A. FOLEY
it is heavily head marking, and, while the morphological structure of words

is very extensively elaborated, the syntactic structure of phrases and clauses
is only weakly so. In fact, Yimas almost entirely lacks the familiar syntactic
category of phrases, and at the clause level it is highly non-configurational
(Hale 1983). The one notable difference of Yimas from most other polysyn-
thetic languages is an extensive system of noun classes, eleven major ones
and half a dozen minor ones. For a detailed exposition of Yimas grammar,
see Foley (1991).
23.2 Nouns, Verbs and Agreement
As a polysynthetic head-marking language, the morphology of agreement is

much in evidence in the language. The language has two major word
classes, nouns and verbs. Words corresponding in meaning to adjectives in
English or other familiar languages are divided between these two classes,
with most belonging to the verb class, though there are three true adjec-
tives. Many grammatical categories distinguish the two major classes of
nouns and verbs, but a quite noticeable contrast is in their formal patterns
of inflection: nouns (and the adjectives) are inflected for noun class and
number by a set of suffixes, while verbs mark the noun class and number of
their core argument participants for the most part with prefixes, as in this
example.
(1) yura antmaŋkl kpa-ŋkl kla-n-am-(n)tut

dog.iii.sg cockatoo.vi.dl big-vi.dl vi.dl.nom-iii.sg.erg-eat-rm.past
‘The dog ate two big (sulphur crested) cockatoos.’
The marking of noun class and number on nouns is largely fusional, so it is

difficult to separate the root from the inflection, and generally both noun
class and number are portmanteau as well (especially in the singular), and
there are morphophonemic rules at work too complex to go into here.
However, the nominal agreement suffixes are isolatable when they appear
on adjectives which are in concord with nouns they modify, as with -ŋkl
vi.dl on the adjective root kpa ‘big’ in (1). Verbs have overt affixes that mark
noun class and number of their core arguments, and these work under a
very complex split ergative-accusative and direct-inverse alignment that will
be explained below, so they are labelled in (1) simply as nominative and
ergative. The distinction between the two major classes, noun and verb, is
very rigid in Yimas; there is no overlap or flexibility so salient in Austrones-
ian languages. A verb can only be used as a noun when it is overtly derived
by a morphological process of nominalization (this will be the focus of the
second half of this chapter), while nouns can never be used as verbs. There
are no processes of denominal verbalization in the language, nothing
equivalent to English I hammered the nail down into the floor, where the noun
hammer denoting an instrument is used as verb denoting doing an action

A Grammatical Sketch of Yimas 863
using that instrument; the equivalent Yimas expression would have to be

something like ‘I hit the nail with a hammer piercing (the floor), going
down.’ Verbs can only be used as arguments when they are derived into
nouns by nominalization. Unmarked predicates are always verbs. The lan-
guage does possess a copula (of very complex morphology showing agree-
ment in noun class and number), and this is required whenever a noun
functions as a predicate.
(2) k-n akrŋ akk

vi.sg-fr.dist frog.vi.sg cop.vi.sg
‘That’s a frog.’ (said on hearing the croaking)
23.3 Clause Structure, Information Structure and

Head Marking
The syntactic templates to build syntactic structures in Yimas are very

simple, in marked contrast to the elaborate morphological templates for
verbs, which given the space limitations, I will not be able to explore in
detail here, though here is a summary of the morphological possibilities for
finite verbs (non-finite verbs are nominalizations); a number of these slots
permit multiple fillers:
MOD-PRO-ASP-ADV-ELEV/DIR-VAL-NINCORP-V-VAL-DIR-ASP-TNS-PC-PRO
As this formula demonstrates, much of this word-level complexity has to

do with the nesting of operators at various levels (Foley 2017b), but here is
a typical example: ka-mpu-pay-ma-takat-ɲa-mpan-m (likely-3pl.erg-first-
inside-touch-imp-3pl.dat-vii.sg.nom) ‘let them first apply it (paint) to them
inside’. Essentially, below the clause level there is not very much syntac-
tic structure; the language strongly prefers to build up clauses simply by
stringing words together, and the ordering of these words within the
clause is highly flexible. While Yimas, like languages of many Papuan
families, has some typological markings of being right-headed, it is by no
means verb-final. Any order of the verb and its arguments and adjuncts is
acceptable, except verb-initial variants. Given this, there is no evidence
for a pre-core slot or a post-core slot, and surprisingly, even the left-
dislocated position seems weakly developed, mainly restricted to vocative
uses, if these are to be analysed as such. The right-dislocated position, by
contrast, is quite often used, essentially for afterthoughts, that is, when
the speaker believes the pronominal affix for a core argument on the verb
is insufficient to identify its referent, and this most commonly occurs
when the referent is a subject, so it somewhat functions then as a switch-
reference device when there are multiple animate participants in an
ongoing text. As a common tendency, also, locative or temporal expres-
sions tend to occur at the periphery of the clause, the beginning or the

end (and again usually the former), but this is only a tendency, and it is
not unusual to find them midway in a clause, say, following a noun
functioning as a core argument.
In any case, Yimas clauses, like those of many Papuan and other purely
oral languages, tend to be information-poor, containing few constituents.
Clauses containing a verb and two overt core argument nouns are very rare
indeed, much less than 1 per cent of all clauses. Where they might occur is
at the beginning of a narrative to introduce the characters. Yimas has no
specialized presentative constructions for introducing new referents like
‘there was an old woman who lived . . .’, so new nouns are introduced baldly
in new mentions at the beginning of a story, and this usually by naming
them; ‘Yapalmay and Mampalmay lived with their brother Yampwiŋkawi’,
for instance, is the opening line of the legend which explains how men’s
penises got shortened. There is no distinct clause type for sentence focus.
About half of all Yimas clauses with transitive verbs consist of just the verb
with perhaps a peripheral temporal or locative expression (often marked
with the oblique suffix -n ~ -nan, the only true nominal case marker in the
language), and the other half will have a single overt noun as a core
argument, again with perhaps a peripheral adjunct. What determines these
patterns is mainly information structure. Nouns whose referents are acti-
vated or easily accessible are normally only expressed by the corresponding
bound affixes for their noun class and number. To have an overt noun and
its corresponding bound affix co-occur in a context in which there is no
ambiguity as to their reference makes the referent of the noun highly
contrastive. It takes on this contrastive reading from the conflict in inter-
pretation between the overt noun and the bound affix. Overt nouns are used
when their referents are inactivated or inaccessible and thereby being
introduced into the stream of speech; they are focal in information struc-
ture, either part of predicate focus or themselves as narrow focus, as clearly
shown in this question-pair.
(3) Q: wara ipa-n(a)-am-n?

what 1pl.nom-pres-eat-pers
‘What are we going to eat?’
A: numpran ipa-n(a)-am-n
pig.iii.sg 1pl.nom-pres-eat-pres
*??numpran na-kay-ɲ(a)-am-n
pig.iii.sg iii.sg.nom-1pl.erg-pres-eat-pres
‘We’re going to eat pork.’
The combination of the overt noun and bound verbal prefix here is very
strange because it forces a contrastive reading where the question does not
set a context in which this is felicitous. The overt noun signals an inactivated
reading, while the bound verbal prefix, an activated or easily accessible one,
two contradictory readings that could only be reconciled in a contrastive
focus reading, but this makes little sense in this context. So the two most

common clause types for Yimas are those with just a verb, where the refer-
ents of all core participants are activated or accessible, or a verb and one
overt noun, in which the overt noun has an inactivated or inaccessible
referent. This is the basic system, though there are a few wrinkles and
complications, particularly with wh-questions (Foley 1991: 430–433).
Due to the sensitivity of the bound verbal affixes to information struc-
ture, the analysis of the head-marking language Lakhota in Van Valin
(2013) needs to be expanded a bit. Van Valin (2013) did not intend his
analysis of Lakhota to be applicable to all head-marking languages, but
rather focused on the type represented by Lakhota, and in this spirit we
offer this analysis of Yimas as a further enrichment of the descriptive
typology. For Lakhota, Van Valin argues that the verbal prefixes, which
may be null, saturate the argument positions within the core and that any
overt nouns are in core-external, but clause-internal positions, essentially
sisters of the core node under the clause. Given that no basic bound affix of
any person, number or noun class combination in Yimas has a null expo-
nent, it would be extremely undesirable to appeal to null affixes in this
language, but without them, to apply Van Valin’s analysis of Lakhota to
Yimas would require violating the Completeness Constraint, an even more
undesirable result, if we want to satisfy all the lexical requirements of the
verb and its subcategorized arguments in the core, as we clearly do in this
language with no voice alternations or extractions. So instead, I propose
that both bound verbal affixes and overt nouns occur in the core, as long as
the nouns are inactivated, that is, they do not co-occur with a bound prefix.
Those that do, occur in the same core-external position that Van Valin
proposes for Lakhota. All of this is illustrated in the following very typical
Yimas sentence and its representation in layered structure and informa-
tion structure. The only constraint on word order here is that the conjunc-
tion kanta ‘but’ occurs preferentially in second position, though this is not
rigid, and the verb cannot be initial; otherwise all possibilities are accept-
able, though some will be more marked pragmatically. Note particularly
the separation by this conjunction of the modifier kamta- ‘empty’, which is
actually a nominalized verb in a relative clause (Section 23.7.2), hence the
tense marker -k irr, from the noun parwa ‘dock’ that it modifies, but linked
to by the noun class and number concordial suffix proper to nouns of this
class and number. Note too the fact that the oblique case suffix -n only
occurs on the noun parwa ‘dock’, but applies to its linked modifier kamta-
‘empty’ as well. The bound verbal affix is a circumfix for this verb because it
has negative polarity (Foley 1991: 251–263).
(4) parwa-n kanta kamta-k-wa ta-pu-tay-kiak-rm kay

dock.ix.sg-obl but empty-irr-ix.sg neg-3-see-irr-3dl.nom canoe.viii.sg
‘But they both didn’t see a canoe at the empty dock.’
In Figure 23.1, which represents the structure of (4), RX stands for referential
expression, in order to treat nouns and bound prefixes with a single label

SENTENCE
CLM CLAUSE
PERIPHERY CORE RXVIII.SG
RXIX.SG.OBL NUCLEUS RXNOM
NUCN PERIPHERYN PRED
RXIX.SG
parwa-n kanta kamta-k-wa ta-pu-tay-kiak-rm kay

dock.IX.SG-OBL but empty-IRR-IX.SG NEG-3-see-IRR-3DL.NOM canoe.VIII.SG
FOCUS
V (INACT)
NUCLEUS
CORE
SPEECH ACT CLAUSE TNS
SENTENCE
Figure 23.1 Clause structure with constituent, operator and focus projections
and avoid positing phrases where there are none, as such are very depaup-
erate in Yimas. I have also simplified for purposes of exposition the
representation of the relative clause, as these will be discussed in more
detail in Section 23.7.2, and treated it like an adjective, as here it behaves
exactly as a true adjective like kpa ‘big’ would.
The syntactic structure of phrases is extremely simple in Yimas, and what
look like noun phrases are more akin to compounds in structure than

phrases. Essentially a noun phrase can consist of no more than two constitu-
ents, a noun and its preposed modifier, and those modifiers can only be of
two kinds, a possessor, which must be a pronominal or a proper name, or
one of the three adjectives, and these cannot co-occur. We can say ama-na kay
1sg-poss canoe ‘my canoe’ and kpa kay big canoe ‘big canoe’ in that fixed
order of constituents, but a noun phrase like *ama-na kpa kay 1sg-poss big
canoe is impossible in Yimas. One would say ama-na kay kpa-y 1sg-poss
canoe.viii.sg big-viii.sg ‘my canoe, the big one’, where the adjective kpa
‘big’ has been converted into a noun by the noun class and number con-
cordial suffix -y and, like kamta-k-wa empty-irr-ix.sg ‘empty’ in (4), can be
floated away from the noun it modifies and no longer forms a constituent
with it, so that this is actually two referential phrases in apposition to each
other. Ignoring nominalizations of verbs to be discussed below, referential
phrases in Yimas can be made up of at most two constituents, such as ama-na
kay or kpa kay, and the vast majority consist of just one, so I will continue to
refer to them as referential expressions (RX) in preference to RPs, especially
as they are typically at the word (X0) level. The only operator that RXs take is
number, at the core level, singular, dual and plural, and adjectives and
possessives are peripheral modifiers, so, as in Figure 23.2.
RX RX
COREN COREN
PERIPHERYN NUCLEUSN NUCLEUSN
ADJ N N
kpa kay kacmpt

big canoe.VIII.SG canoe.VIII.PL
NUCLEUSN NUCLEUSN
COREN NUM COREN NUM
RX RX
Figure 23.2 Phrase structure with constituent and operator projections

The only other phrase type in Yimas is the postpositional phrase headed by
postpositions like nampan ‘toward, for, because of’ or kantk ‘together with’,
which form the predicate and nucleus of the phrase and their core argu-
ments are RXs. Postpositional phrases are always predicative peripheral
adjuncts; there are no non-predicative peripheral adjuncts in the language,
because, as we shall see next, all arguments subcategorized by a verb must
be projected as arguments in the core, either as an independent RX or as a
bound verbal one.
23.4 Logical Structures and Macroroles
Yimas, like some other Papuan languages (e.g. Kalam; Pawley 1993), has a
restricted lexicon of monomorphemic verbs, only around a couple of hun-
dred verb roots. More complex verbal expressions are constructed by these
roots in serial verb constructions in nuclear junctures. Yimas has basic
underived intransitive, transitive and ditransitive verbs, although there
are only four roots in the last class. There are also lexical processes that
derive verbs, a reciprocal prefix that removes a core argument and two
causative prefixes, six applicative affixes and possessor raising that add
one, but unlike some Bantu languages such as KiHaya (Duranti and Byar-
ushengo 1977), in no case can a verb be derived that exceeds the number of
arguments of an underived verb; the maximum number of core arguments
of any verb is three. Beyond these processes of decreased and increased
transitivity, there are no derivational processes in Yimas that affect a verb
and its arguments’ mapping into constituent structure. There are no voice
alternations such as passive or antipassive and no alternations in realization
such as we find in English between the farmer loaded hay on the truck versus the
farmer loaded the truck with hay. Given a verb’s logical structure and the
arguments it subcategorizes, there is one and only one mapping into con-
stituent structure that corresponds to it. The case marking of the arguments
is very complex and will be discussed in the following section; here I am just
concerned with their basic realization as constituents. Consider an intransi-
tive unaccusative verb like mal- ‘die’, as in na-mal 3sg.nom-die ‘he just died’.
This is an achievement predicate and so has the following logical structure:
INGR died′ (he). As he is an argument in a frame of pred′ (x), it will be
assigned the undergoer macrorole by the universal Actor-Undergoer Hier-
archy (AUH), and as the sole argument will appear in nominative case as a
bound prefix to the verb. Now consider an intransitive unergative verb like
iray ‘cry’ as in na-iray 3sg.nom-cry ‘she cried’. This is an activity predicate
with the logical structure do′ (she, [cry′ (she)]). As she is an argument in a
frame of do′ (x), by the universal AUH it will be assigned the actor macro-
role, but again as the sole argument will appear in nominative case as a

bound prefix to the verb. Now consider a canonical transitive verb such as
warapa- ‘cut with a flat sharp instrument’ as in (5).
(5) kaŋk-ɲan na-ka-warapa-ntut

shell.vi.sg-obl 3sg.nom-1sg.erg-cut-rm.past
‘I cut it with a shell.’
The logical structure of this clause, including the oblique non-

subcategorized RX, would be as in (6).
(6) [do′ (I, Ø)] CAUSE [[ do′ (shell, Ø)] CAUSE [INGR cut′ (it)]]
By the universal AUH, it as the argument of pred′ (x) would be assigned the
undergoer macrorole and therefore be realized by nominative case in this
direct arrangement of arguments (see Section 23.5). But there are two
potential actors, I and shell, as both are first arguments of do′ (x, . . .) (Yimas,
unlike Tagalog, has no constructions which differentiate grammatically
between volitional deliberate actions and those which are not, and there
are no verb roots which lexicalize this contrast like see versus watch, so there
is no need to postulate an operator DO in logical structures.) Yimas, unlike
English, does not permit sentences like the shell cut it; the animate instigator
who instigates the causal chain must be mentioned overtly and outranks the
instrument. Hence it is the animate instigator, here I, which is assigned the
actor macrorole and receives ergative case, realized in the core along with
the undergoer, while the instrument is realized as an oblique RX in the
periphery of the core. Finally, consider the canonical ditransitive verb ŋa-
‘give’ as in (7).
(7) trawsistm tma-mpu-ŋa-ŋa-ntut

trousers.v.dl v.dl.nom-3pl.erg-1sg.acc-give-rm.past
‘They gave me two pairs of trousers.’
Note that Yimas belongs to the relatively rare type of triple-agreement

languages; given the proper information structure configuration, all three
arguments subcategorized by the verb are realized through bound pronom-
inal affixes on the verb, and so all three must be regarded as core arguments.
Yimas seems to be a mix of a direct object language or a primary object
language, but just what is it? There are two non-actor core arguments, the
theme and the recipient, accessible to the undergoer macrorole, but which
one or is there only one? The relevant principles in the revised AUH given in
Van Valin (2005: 126) are Principle A: lowest-ranking argument in argument
structure or Principle B: second-highest-ranking argument in argument
structure. Consider (8), the logical structure for (7).
(8) [do′ (they, Ø)] CAUSE [BECOME have′ (me, trousers)]
The actor macrorole is straightforward: as they is the first argument of do′

(x, . . .) and thereby is assigned the actor macrorole and realized in ergative

case. But the undergoer is more problematic: by Principle A, trousers, the

lowest-ranking argument would be the undergoer, but by Principle B, it
would be me, the second-highest-ranking argument. Is there any way to
determine which or must both be reckoned as undergoers? The answer to
this question is not easily apparent. There are no voice alternations to check
for accessibility, and the evidence is conflicting, but the bulk seems to
favour the claim of two undergoers, as in Lake Bantu languages (Bresnan
and Moshi 1990; Duranti and Byarushengo 1977). For example, the case
hierarchy to be described in Section 23.5 mandates the following ranking
ACC > ERG > NOM. Note in (7) that the referent of trousers is indicated by
nominative case and me by accusative. This might establish me as the real
undergoer as it outranks trousers in this case hierarchy. But as we shall see,
nominative is usually the only obligatory case on a verb if it has any
agreement prefixes at all, so it would seem strange to claim less obligatory
affixes as marking the undergoer over more obligatory ones; this favours
trousers as the undergoer. Yet the nominative case affix always occurs on the
periphery of the verb, either right at the beginning or right at the end, in
contrast to the more highly salient internal positions closer to the verb
stem occupied by the recipient affix. This, then, again may favour the claim
of me as the undergoer. On the other hand, recipient arguments cannot be
directly relativized like the actor or theme, but require structures that are
ambiguous with coordinated clauses; the fact that the actor and theme
pattern alike as against the recipient would seem to favour trousers as the
undergoer. Yet in possessive-raising constructions like ‘he hit me on the
arm’, for which in head-marking languages the affected person is cross-
linguistically typically the undergoer, not the body part, the affected
person in Yimas is always expressed exactly as the recipient of a ditransitive
verb (Foley 1991: 300–303) (for comparable data, see the Muskogean
languages, Choctaw (Broadwell 2006), Chickasaw (Munro 1984; Munro and
Gordon 1982) and Creek (Martin 2011)). The evidence is murky and incon-
clusive, but tentatively, I conclude that ditransitive verbs in Yimas take two
undergoers, one by Principle A and one by Principle B. But paralleling the
way RRG has rejected the notion of a universal grammatical category of
subject, although it can serve as a useful descriptive category for some
languages such as English, perhaps we should consider abandoning the
concept of the macrorole undergoer as a necessary universal category
present in the grammar of all languages, although again valuable for many
languages. The concept of undergoer does no necessary descriptive work in
Yimas, and insisting that we find one raises a host of problems. The lack of
verb-based lexical alternations as with English load above also suggests its
lack. The behaviour of applicative affixes in the language can be simply
described in terms of adding core arguments with no reference to the
concept of undergoer, which in fact just complicates their description.
But in keeping with the spirit of RRG and this volume, I will use the notion
in the remainder of this chapter. The question now advanced is whether

Yimas is unique or do other triple-agreement languages pose the same

challenge? It appears at least the Muskogean languages do.
23.5 Case Marking: Split Ergativity and Inversion
Case marking in Yimas is very complex, but there is no quirky or semantic-

ally based case. All core case marking is determined by the function and
relative topicality of the arguments, and all core case is realized by verbal
bound pronominal affixes. The only other case marking is the sole -n ~ -nan
for non-core constituents illustrated in (4) and (5). Core case marking in
Yimas is a complex blend of split ergative-accusative alignments overlaid by
a direct-inverse system. The split system separates the local persons, namely
the immediate speech act persons, first and second, from the non-local
person, the third person. The local persons, first and second, align according
to a three-way system, distinguishing the transitive actor in ergative case,
from the intransitive subject, actor or undergoer, S in Dixon’s (1979) terms,
in nominative case, and from the transitive undergoer in accusative case.
The non-local third person distinguishes the transitive actor with ergative
case from the intransitive subject and the transitive undergoer, both
marked with nominative case. Tables 23.1 and 23.2 present the forms.
Note that the second-person and third-person singular ergative prefixes
are homophonous. For ditransitive verbs, the theme argument, that which is
given or transferred, is indicated by the relevant nominative prefix, and the
recipient by the accusative prefixes for the local persons, but by a distinct set
of dative suffixes for non-local persons. In the actual marking of verbs with
Table 23.1 Yimas agreement affixes for local persons
PRONOUN ERG NOM ACC
DL kapa ŋkra- kapa- ŋkra-

1 PL ipa kay- ipa- kra-
SG ama ka- ama- ŋa-
DL kapwa ŋkran- kapwa- ŋkul-
2 PL ipwa nan- ipwa- kul-
SG mi n- ma- nan-
Table 23.2 Yimas agreement affixes for non-local persons
PRONOUN ERG NOM DAT
SG mn n- na- -nakn
DL mrm mp- impa- -mpn
PC mŋkt ŋkl- kra- -ŋkt
PL mum mpu- pu- -mpun

these affixes, two principles apply, both aspects of the direct-inverse system
of the language: first, local persons outrank non-local persons, specifically
1st > 2nd > 3rd, and second, surprisingly, accusative prefixes outrank
ergative ones which in turn outrank nominative ones. There are also two
purely structural constraints: one, the most highly ranked prefix must
appear in the immediately preverbal position, and, two, every verb should
have a nominative affix, and that must occur on the left edge of the verb in
initial position, unless usurped or demoted by a modal or other inflection to
the right edge, as in (4). Let’s see how this system all works. Consider how
one would say ‘I hit them’ versus ‘they hit me’ in Yimas. The direct form, ‘I
hit them’ (local person acting on non-local person) is straightforward. The
local person is the transitive actor and ergative, hence ka- 1sg erg. The non-
local person is the transitive undergoer, and the case-marking system for the
non-local person is ergative-nominative, so the undergoer is realized as
nominative, pu- 3pl nom, simultaneously satisfying the requirement for
an overt nominative. Because local > non-local and ERG > NOM, the first-
person ergative prefix will occupy the salient immediately preverbal pos-
ition, and the third-person nominative the left edge.
(9) pu-ka-tpul
3pl.nom-1sg.erg-hit
‘I hit them.’
The inverse form ‘they hit me’ is a little more complicated. The undergoer
is a local person, and local persons have accusative forms (e.g. ŋa- 1sg acc).
Note that the local person is higher by both local > non-local, and ACC >
ERG, so ŋa- 1sg acc must occupy the salient immediately preverbal position.
The non-local actor would normally take ergative case, and indeed there is
an ergative form for third plural: mpu- 3pl erg. But the expected form *mpu-
ŋa-tpul is ungrammatical because it runs afoul of the requirement for an
overt nominative; such a verb lacks a nominative prefix on the left edge. So
instead, the third plural pronominal is realized by the corresponding nom-
inative prefix, pu- 3pl nom.
(10) pu-ŋa-tpul
3pl.nom-1sg.acc-hit
‘They hit me.’
Ditransitive verbs are no different. Consider ‘I gave it to them’ versus ‘they

gave it to me’, the first, direct, the second, inverse. The direct form is
straightforward. The non-local recipient undergoer is realized by a dative
suffix, so it doesn’t compete with the interaction of the prefixes where the
direct-inverse system holds sway. The actor, the ergative pronominal, is first
person and therefore the higher ranked by both local > non-local and ERG >
NOM, so it appears in the immediately preverbal position. The theme argu-
ment undergoer is inanimate; themes with human or higher animate

referents are completely prohibited for ditransitive verbs in Yimas. The

affixes in Tables 23.1 and 23.2 only apply to referential expressions with
human or very high animate referents, like pigs, dogs or crocodiles. Nouns
with such referents belong to one of three noun classes. Nouns with all other
referents and particularly all nouns with inanimate referents belong to one
of the other seven major or a handful of minor noun classes. These classes
also have verbal bound referential expressions, but they usually occur in
only one case, nominative, again demonstrating the privilege of the nom-
inative case. For our examples here let us take muraŋ ‘oar’, a noun of noun
class VI, as the referent of ‘it’, the theme undergoer; the relevant nominative
verbal affix is k- vi.sg.nom. As it is a nominative prefix, it satisfies the
requirement of an overt nominative and will appear on the left edge, so
the resulting form is as in (11).
(11) k-ka-ŋa-r-mpun
vi.sg.nom-1sg.erg-give-perf-3pl.dat
‘I gave it to them.’
The inverse form is more interesting, as local recipient undergoers are

realized by accusative pronominals, the most highly ranked. As a first-
person undergoer realized by an accusative pronominal, it ranks highest
by both local > non-local and ACC > ERG > NOM, so it must occupy the
immediately preverbal position. Now the third-person plural actor and the
third-person singular theme undergoer slug it out. Both are third person,
so local > non-local will not distinguish them, but ERG > NOM will. The
actor argument is ergative, and because the theme undergoer is inanimate,
it only has the option of being nominative, so by ERG > NOM, the actor
argument is of the next highest rank and occupies the position next closest
to the verb after the accusative prefix, while the hapless theme argument
satisfies the requirement of an overt nominative on the left edge.
(12) k-mpu-ŋa-ŋa-t
vi.sg.nom-3pl.erg-1sg.acc-give-perf
‘They gave it to me.’
Let’s see how we could handle this very complex case system by extending
to Yimas case assignment the rules for ergative/accusative constructions
provided in Van Valin (2005: 108), reproduced here as (13). (There is a much
more detailed and typologically varied description of these in Van Valin and
La Polla (1997), but the points made below would still apply.)
(13) Rules for accusative case constructions:

a. assign nominative case to the highest-ranking macrorole
b. assign accusative case to the other macrorole
Rules for ergative case constructions:
a. assign absolutive case to the lowest-ranking macrorole
b. assign ergative case to the other macrorole

It is clear that these will need some modification for Yimas. First, we
have no need for an absolutive case, and invoking it would greatly
complicate the description. So-called absolutive case is simply nomina-
tive, in Yimas and, I would argue, universally. That is why, like nomina-
tive in accusative constructions, erstwhile absolutive case constituents
are normally formally unmarked and why the nominative is always the
privileged case in Yimas regardless of whether it shows up in the
accusatively aligned local persons or the ergatively aligned non-local
persons. Further, the problem of the lower-ranked macrorole in (13a) is
problematic in the case of ditransitive verbs with two undergoers, but
let us see how we go. For non-local persons the modification is very
straightforward: simply assign nominative case to the lowest argument
on the AUH (and dative case to the other one if present) and ergative
case to the highest macrorole. The rules for case assignment to the local
persons are more complex, because the rules for accusative and ergative
case constructions apply simultaneously and partially, that is, only
something like the (b) conditions apply. If we were to apply all the
conditions, we will get case conflict, a transitive actor will be assigned
nominative case by accusative condition (a), but ergative by ergative
condition (b). But if we absorb the (a) conditions into and restate the
(b) conditions, the system will work: (a) assign accusative case to the
lower-ranked macrorole when there are at least two (there is a compli-
cation with ditransitive verbs with local recipients; there assign accusa-
tive case with the next to highest (or lowest, either will work) argument
on the AUH); (b) assign ergative case to the highest-ranked macrorole,
again where there are at least two; (c) assign nominative to any core
argument that has not received case by (a) or (b). The tree diagram in
Figure 23.3 represents the mapping of (11) using these rules from logical
structure to constituent structure.
23.6 Pivots or Privileged Syntactic Arguments
There is no evidence in Yimas for a privileged syntactic argument

(PSA) such as subject in English or the ang phrase in Tagalog. Different
constructions target different constituent types, and no one type
emerges as privileged across the wide range of constructions. Most
constructions in Yimas simply target the notion of core argument.
Yimas, like many Papuan languages, employs clause chaining as a
clause linkage device (see Roberts, this volume, Chapter 25). But unlike
most, it has no restrictions for this to shared subjects or switch-
reference tracking and it permits clause chains to be formed on any
shared argument without any morphological difference, including
none (14b).

SENTENCE
CLAUSE
CORE
RX RX NUCLEUS RX
PRED
k- ka- a-r- -mpun

VI.SG.NOM 1SG.ERG give-PERF 3PL.DAT
CASE: ERG DAT NOM
MACROROLES: ACTOR UNDERGOER UNDERGOER
LEXICON: [do′ (I, Ø)] CAUSE [BECOME have′ (them, it (VI.SG))]
Figure 23.3 Linking from semantics to syntax in (11) of Yimas
(14) a. tmal kray-mpi ya-kay-am-wat amtra

sun.v.sg dry-seq V.pl.nom-1pl.erg-eat-hab food.v.pl
‘the sun having dried it, we eat the food’
b. tmal l-ŋka-p(u)-mpi kumpwia mnta
sun.v.sg down-go-away-seq flying.fox.viii.pl then
wa-kay-tay
viii.pl.nom-1pl.erg-see
‘The sun having set, then I saw flying foxes.’
In (14a) the argument shared between the clauses and elided from the first
clause is the undergoer, while in (14b) there are no shared arguments. Some

constructions do have an intransitive subject/transitive subject (S/A) pivot,

such as genitivization in nominalized non-finite complements.
(15) a. God-na anti papak-t-wal

God-poss ground.viii.sg carve-nfn-custom.v.sg
‘God’s making of the world’
b. *anti-ɲa God papk-t-wal
ground.viii.sg-poss God carve-nfn-custom.v.sg
‘the world’s making by God’
c. Yakayapa(n)-na am-t-wal
pn-poss eat-nfn-custom.v.sg
‘Yakayapan’s (way of ) eating’
In (15a) the actor of a transitive verb papk ‘carve’ has been genitivized in the
nominalized non-finite complement, but if we try this with the undergoer in
(15b), the result is ungrammatical (though okay in English). However, with
the single argument of an intransitive verb (15c), genitivization is again
possible. Other constructions, though, work on an intransitive subject/tran-
sitive object (S/O) pivot, for example the scope of elevational/directional
affixes. Consider these examples.
(16) a. kay i-ɲa-l-ampu-n
canoe.viii.sg viii.sg.nom-pres-down-float-pres
‘The canoe is down there.’
b. kay naŋ-l-arm-na-ŋkan-i
canoe.viii.sg imp.pl-down-board-imp-pc-viii.sg.nom
‘You all board the canoe down below.’
‘*You all down below board the canoe.’
Note that in (16a) the scope of the elevational prefix l- is over the sole
argument of the intransitive verb ampu- ‘float’, while in (16b) it is over the
undergoer of the transitive verb arm- ‘board’ and cannot be construed as
modifying the actor, a clear S/O pivot. The only way to say the equivalent of
the starred translation would be something like ‘you all are standing/sitting
down below and now you all board the canoe’.
23.7 Clause Linkage and Nominalization
23.7.1 Non-Finite Nominalizations

Yimas contrasts with many Papuan languages and in particular with those
of the Trans New Guinea family like Amele (Roberts, this volume, Chapter 25)
in preferring subordination rather than the cosubordination by extensive
clause chaining as its favourite clause linkage device. It accomplishes
subordination through nominalizations, both non-finite and finite. Yimas
completely lacks the category complementizer; all nominalizations are
simply complex nouns, and as nouns, all must be marked for noun class
and, if relevant, number. The formal template is easiest to see with non-
finite agentive nominalizations, such as irut ampa-r-mprum mat.ix.pl weave-
nfn-ii.dl ‘two female mat-weavers’; a parallel full finite clause would be:

irut imp(a)-ampa-wat mat.ix.pl 3dl.nom-weave-hab ‘they both weave mats.’

Note first of all that the nominalization makes a distinction in noun class,
here noun class II, the class for female humans, that the corresponding
finite verbal prefix impa- 3dl.nom cannot; this is in keeping with its status
as a noun, distinguishing the full range of classes that nouns do. The finite
tense marker -wat hab is replaced by the non-finite marker -ru (with a
number of allomorphs). All core argument marking by bound verbal affixa-
tion beyond the realization of the agent that the entire nominalization
refers to is now completely prohibited; the undergoer of the finite clause
can be carried over into the nominalization, but it cannot ever be realized by
a bound RX: *irut w(a)-ampa-r-mprum mat.ix.pl ix.pl.nom-weave-nfn-ii.dl,
*w(a)-ampa-r-mprum ix.pl.nom-weave-nfn-ii.dl. Nor can the undergoer be
genitivized as a modifier of the nominalization, as it can in English: *irut-
ɲa ampa-r-mprum mat.ix.pl-poss weave-nfn-ii.dl ‘the two female weavers of
mats’. This indicates that the undergoer must remain inside the core of the
nominalization to satisfy the Completeness Constraint for the verb. But the
whole nominalization can be modified, for example by an adjective: irut
ampa-r-mprum yua-mprum mat.ix.pl weave-nfn-ii.dl good-ii.dl ‘two good
female mat-weavers’. The concordial pattern here tells us that the head of
the nominalization is the noun class and number suffix. The non-finite verb
is embedded underneath it as a core, with its subcategorized arguments.
Because this is an agentive nominalization, the actor is actually bound by
the head of the nominalization, the noun class and number suffix, and
therefore prohibited from occurring. This all suggests the structure in
Figure 23.4 for this construction.
All non-finite nominalizations in Yimas have the structure of Figure 23.4.
As we shall see, the structure of finite ones is very similar except that the
node of clause is introduced with its operators, modifiers and inflectional
possibilities. The nucleus of a non-finite nominalization can be as complex as
a finite verb, simply lacking tense and verbal pronominal agreement, and
can consist of incorporated adverbials, applicative derivations or verbs in a
serial verb construction, that is, verbs linked in a nuclear juncture, as in
these examples: kpa-nti-pramuŋ-tu-mat big-adv-sleep-nfn-i.pl ‘men who sleep
too much’, taŋkway-cakal-cu-mprum watch.over-feel-nfn-ii.dl ‘the two women
looking after (the kids)’, namtamparawt-ɲan api-c-awt foot.ix.dl-obl put.in-
nfn-sg ‘a sock (that which puts inside at the feet)’. The different kinds of
non-finite nominalizations in Yimas cover the midrange of the Interclausal
Semantic Relations Hierarchy (Van Valin 2005: 208), psych action, jussives,
etc. What distinguishes the different types of non-finite nominalizations
are the different heads or noun class markers they take (or head noun, as
in complements of desire), but they all have the structure of Figure 23.4.
When they function as complements, that is, as core arguments of a
matrix verb, they can and usually do trigger agreement with that verb,
the noun class marker of the nominalization matching the affix on
the verb. Beside agentive nominalizations, there are four other types
of nominalization that function as complements, exemplified in (17).

RX
COREN
CORE NUCLEUSN
RX NUCLEUS RX
PRED
V N
irut ampa-r -mprum

mat.IX.PL weave-NFN II.DL
NUCLEUSN
COREN NUM
RX
Figure 23.4 Constituent structure of a non-finite nominalization in Yimas
These are marked by specific noun class marker suffixes utilized only for
this function, or in the case of complements of desire by a bound noun
wampuŋ, literally ‘heart’. Furthermore, as true complements they function
as core arguments of the complement-taking verb of the core of the main
clause and are therefore commonly (17a, b, c), though not obligatorily
(17d), indicated there by a corresponding bound verbal affix of the same
noun class.
(17) a. complement of words or thought (thinking is internal speech)

patn wayk-r-mpwi pia-ka-i-c-mpun
betelnut.v.sg buy-nfn-talk talk.nom-1sg.erg-tell-perf-3pl.dat
‘I told them to buy betelnut.’

b. complement of activity
nam wark-t-nti tia-ka-ira-karŋkra-t
house enclose-nfn-act act.nom-1SG.erg-appl-tired-perf
‘I’m tired of building houses.’
c. complement of customary or habitual action
yaki am-t-wal ntak-na-k
tobacco.v.pl eat-nfn-custom.v.sg leave-imp-V.sg.nom
‘Stop smoking!’
d. complement of desire
tpuk am-t-wampuŋ kpa-n ama-na-t-n
sago.x.sg eat-nfn-heart.v.sg big-v.sg 1sg.nom-pres-feel-pres
‘I really want to eat sago.’ (‘I have a big desire to eat sago’)
Note in (17d) that the adjective kpa ‘big’ modifies the nominalization.
Core arguments of non-finite nominalizations may be the pivot of control
constructions and when they are, the pivot is strictly S/A; undergoers may
never be controlled. Otherwise, genitivization applies, again strictly under
an S/A pivot, illustrated above in (15) and in this example.
(18) patn kpuc-t-wal mpu-na-kn

betelnut.v.sg chew-nfn-custom.v.sg 3pl-poss-v.sg
‘their (manner of ) chewing betelnut’
The controllers of the S/A pivot in control constructions can be any of the
three core arguments, and what determines which is the Theory of
Obligatory Control (Foley and Van Valin 1984: 307–311; Van Valin 2005:
243): (1) causative and jussive verbs have undergoer control and (2) all
other transitive verbs have actor control. In control structures in Yimas
the nominalized complement always functions as an undergoer, so with
simple transitive complements there is no issue: the controller can only
be the actor, the only other macrorole, as in (17b, d) above and in the
examples below.
(19) a. patn wayk-r-mpwi pia-ka-kacapal

betelnut.v.sg buy-nfn-talk talk.nom-1sg.erg-forget
‘I forgot to buy betelnut.’
b. yaki am-t-wampuŋ na-na-t-n
tobacco.v.pl eat-nfn-heart.v.sg 3sg.nom-pres-feel-pres
‘He wants to smoke.’
Ditransitive verbs are more complicated as there are two potential con-
trollers, the actor or the second undergoer. In accord with the Theory of
Obligatory Control, if the complement is a causative or jussive one, there is
undergoer control, as in (17a) and (20).
(20) impram pay-c-mpwi na-kra-kankantakal

basket.vii.sg carry-nfn-talk 3sg.nom-1pl.acc-ask
‘He asked us to carry a basket.’

Note that crucially the jussive or causative semantics is necessary; just

because a verb of saying is ditransitive does not mean it will have undergoer
control.
(21) kay yamal-c-mpwi pia-mpu-ŋa-taŋ-tmi

canoe.viii.sg carve-nfn-talk talk.nom-3pl.erg-1sg.acc-appl-talk
‘They talked with me about (my, their or someone else’s) building a canoe.’
Here, although the verb is ditransitive through derivation with an

applicative, there is no obligatory control, because the semantics is not
causative or jussive. Ditransitive verbs that do not have causative or jussive
semantics can have, as expected, actor controllers.
(22) tpuk am-t-mpwi pia-mpu-ŋa-taŋkway-cmi

sago.x.sg eat-nfn-talk talk.nom-3pl.erg-1sg.acc-appl-talk
‘He told me (while looking at me) about (his) eating sago.’
In addition to functioning as complements, non-finite nominalizations

can function as peripheral adjuncts and as modifiers, essentially non-finite
relative clauses. Let us look at the latter first as they are simpler. All relative
clauses in Yimas, finite and non-finite, are simply juxtaposed in apposition
to the noun or referring expression they modify, just as adjectives can be
(17d), They are adjoined (Hale 1976) and not embedded, as in this example.
(23) pu-k namat pu-ŋkl-awl-k

3pl-prox person.i.pl 3pl.nom-3pc.erg-get-irr
[nampt wark-r-mat]
house.pl enclose-nfn-i.pl
‘They few got these people who build houses.’
The structure of (23) is essentially that of a non-finite nominalization nampt

wark-r-mat house.pl enclose-nfn-i.pl ‘the ones (masculine) building houses’
as in Figure 23.4 modifying a core argument pu-k namat 3pl-prox man.i.pl
‘these men’, in the same way as kamta-k-wa empty-irr-ix.sg ‘empty’ modifies
parwa-n dock.ix.sg-obl ‘at the dock’ in Figure 23.1. Non-finite relative
clauses differ from the nominalizations that we have seen thus far only in
the noun class marker they take. Instead of the specialized noun class
markers for complements in (17), non-finite relative clauses just use the
basic adjectival concord set for the various noun classes and numbers of the
nouns the modify: awruk [awt yara-t-uŋ] torch.x.sg [fire yara-nfn-x.sg] ‘a
torch for picking up fire’, with just some exceptions for those modifying
nouns denoting human males or females.
Non-finite nominalizations functioning as peripheral adjuncts always
occur with the oblique suffix -n ~ -nan. They have a very specialized semantic
function, indicating two events which are simultaneous. Such a semantic
relation is quite low on the Interclausal Semantic Relations Hierarchy (Van
Valin 2005: 208), lower, for example, than circumstances, reasons, condi-
tionals and concessives, but this is inaccurate for Yimas, as all of these latter
are expressed by finite nominalizations, while simultaneous events must be

expressed with non-finite nominalizations. While oblique non-finite nomin-

alizations have final noun class and number markers as their heads, require
-ru and have the basic structure of Figure 23.4, they differ in two crucial
respects: first, their actors, if overt (and they are always actors, as I have no
examples of states, achievements and accomplishments occurring in this
construction), must occur as bare RXs like undergoers and may not be
genitivized (24c). In this feature, oblique non-finite nominalizations are
more like clauses than the ones previously presented, though formally they
are still clearly cores attached to a nominal nucleus which is a noun class
and number suffix, which is then inflected for oblique case. And second, the
noun class and number suffixes do not show concord with an external RX,
but rather mark those features for the actor of the nominalization, as in
these examples.
(24) a. [mpa irm-kia-r-ŋkt-ɲan]

now stand-night-nfn-pc-obl
paŋkra-na-ma-ŋka-pu-kia-k
1pc.nom-pres-in-go.by.land-away-night-irr
‘Standing now, we few walk inside.’
b. [wark-r-mat-ɲan nam] kumpwi mnta
enclose-nfn-i.pl-obl house boy.I.pl then
numa-mpu-ntak-t
village.nom-3pl.erg-leave-perf
‘While building a house, the boys left the village.’
c. arm nampt ya-mpu-tawɲcak-kia-k
water house.pl house.pl.nom-3pl.erg-flood-night-irr
[m-um pay-kia-r-mat-ɲan num-un-mat]
nr.dist-i.pl lie-night-nfn-i.pl-obl village-obl-i.pl
‘The water flooded the houses, while they, the villagers, slept.’
In (24a, b) the actor of the nominalizations is the same as that of the matrix
verb, and being such, it can be elided. But such control is not obligatory, as
(24c) demonstrates. There the actor of the nominalization is num-un-mat
village-obl-i.pl ‘villagers’, while that of the matrix verb is arm ‘water’, which
is always formally plural in Yimas, hence the bound prefix mpu- 3pl.erg. Any
core argument can be the controller of an elided actor of an oblique non-
finite nominalization.
(25) pu-kra-ant-t pan-t-mat-nan

3pl.nom-1pl.acc-hear-perf pound.sago-nfn-i.pl-obl
‘They heard us when pounding sago.’ (Either they or we could be pounding
sago.)
Finally, these oblique non-finite nominalizations can be used to describe

locations. In that case, the noun class and number marker is invariably -a ix.
sg, but there is no known synchronic noun of that class in Yimas that the
affix could refer to. It seems to be fossilized, but the overall structure is typical
of a non-finite nominalization: mawrun tu-r-a-n enemy.i.sg kill-nfn-ix.sg-obl

‘a place for killing the enemy, battlefield’, awŋkw-cpaŋ-t-a-n go.down.into.

water-bathe-nfn-ix.sg-obl ‘a bathing place’.
23.7.2 Finite Nominalizations: Relative Clauses

Finite nominalizations occur with tense suffixes and bound verbal pro-
nominal affixes for core arguments. Finite nominalizations are used in
relative clauses and oblique (i.e. adverbial-type) clauses as peripheral
adjuncts of the matrix clause, but they may never be used as comple-
ments (i.e. core arguments of the matrix verb). All finite nominalizations
essentially have the structure of relative clauses, and the formation of
relative clauses is somewhat complicated in Yimas, so I can only give a
brief overview here. Their formation is complicated as they often involve
yet another linear rearrangement of bound pronominal affixes for core
arguments, the complexities of which, due to the split case system and
the direct-inverse system, are considerable. Relative clauses are adjoined,
not embedded, so the noun being modified does not form a constituent
with it and, if recoverable from context, can be elided. The relative
clause itself is headed by a noun class and number suffix on its verb
that is in concord with the noun that it modifies or, if missing, that is
recoverable from the context. This bound noun class and number suffix
indexes one of the core arguments of the verb, actor or undergoer, that
is the relativized noun, and in that sense, relative clauses in Yimas can
be seen as internally headed, as the external modified noun, if present,
is always resumed in the relative clause by a bound verbal affix
according to its noun class and number. That noun class and number
suffix, like those in non-finite nominalizations, occurs at the end of the
verb of the relative clause and heads it, though not necessarily at the
end of the relative clause, as constituents, especially peripheral adjuncts,
but not restricted to them, can follow the verb. The noun class and
number marker binds the missing relativized noun in the relative clause
and provides its referent. The very simplest relative clauses are just of
this structure; generally they do not co-occur with a noun and so
function as RXs on their own: mal-k-n die-irr-i.sg ‘a corpse’, ŋa-t-ø give-
perf-v.sg ‘a gift’, kalc-k-n strengthen-irr-i.sg ‘a strong person’, wa-kia-k-ra
go-near-irr-v.pl ‘means/route of going’. Their structure is represented in
Figure 23.5. These simple relative clauses function as kinds of patientive
nominalizations in opposition to the agentive nominalizations described
earlier.
(26) tu-r-awt tu-t-Ø

kill-nfn-i.sg kill-perf-i.sg
‘killer’ ‘someone killed’
amp-r-awt amp-(r)-ra
kindle-nfn-i.sg kindle-perf-v.pl
‘fire lighter’ ‘firewood’

kalc-r-awt kalc-k-n
strengthen-nfn-i.sg strengthen-irr-i.sg
‘hard worker’ ‘a strong person’
am-t-awt am-t-ra
eat-nfn-i.sg eat-perf-v.pl
‘eater’ ‘food’
Most relative clauses are rather more elaborate than this. The verb of the
relative clause (and many are just the verb) is typically suffixed as a whole to
the near distal deictic stem m- ‘that’. Further, they can have bound
RX
COREN
CLAUSE NUCLEUSN
CORE
NUCLEUS RX
PRED
V N
wa-kia-k -ra
go-NEAR-IRR V.PL
NUCLEUS NUCLEUSN
CORE COREN NUM
TENSE CLAUSE RX
Figure 23.5 Constituent structure of a simple finite nominalization in Yimas

pronominal affixes for their subcategorized core arguments. Consider these

examples.
(27) a. namat [m-na-taw-nt-um mnti]

person.i.pl nr.dist-pres-sit-pres-i.pl there
‘the people who are sitting there’
b. namat [m-kra-tpul-c-um]
person.i.pl nr.dist-1pl.acc-hit-perf-I.pl
‘the people who hit us’
c. namat [m-kay-tpul-c-um]
person.i.pl nr.dist-1pl.erg-hit-perf-i.pl
‘the people who we hit’
Example (27a) illustrates a case where the head and relativized argument is
the intransitive subject. This single core argument is realized solely by the
head suffix, the verb bearing no other bound pronominal bound affix. In
(27b) the relativized noun and head is the actor of a transitive verb, realized
by the final head suffix, while the undergoer is realized by its proper prefix.
Finally, in (27c) the relativized noun and referent of the final head suffix
corresponds to the undergoer argument of a transitive verb, while the actor
appears now as a prefix in its proper position.
All types of core arguments and even peripheral adjuncts are relativizable
(though recipient undergoers with difficulty due to morphological compli-
cations (see Foley 1991: 417–418)); here are some examples.
(28) a. on actor
Elias [m-kra-pay-pra-kia-ntuk-ŋkt-ø mota-nan]
pn nr.dist-1pl.acc-carry-toward-near-rm.past-pc-i.sg motor-obl
‘(It was) Elias who brought us few by motor.’
b. on undergoer
anti [God m-n-papk-ntuk-i]
ground.viii.sg God nr.dist-3sg.erg-carve-rm.past-viii.sg
‘the world which God made’
panmal m-n-tpul-c-(n)ak narmaŋ
man.i.sg nr.dist-3sg.erg-hit-perf-i.sg woman.ii.sg
‘the man which the woman hit’
c. on theme undergoer
impram [m-(n)-nan-(n)a-ampa-ŋa-nt-m]
basket.vii.sg nr.dist-3sg.erg-2pl.acc-pres-weave-appl-pres-vii.sg
‘the basket which she is weaving for you’
d. on peripheral location (again the fossilized suffix -a ix.sg is employed)
[maramara m-mpu-t-r-a-n]
goods.v.pl nr.dist-3pl.erg-lay.down-perf-ix.sg-obl
m-ra ya-kay-ɲa-tacay-kulanaŋ
nr.dist-v.pl v.pl.nom-1pl.erg-pres-see (red: tay-)-walk
‘Where they laid out the goods, we walked around looking at them.’
(Literally ‘at the place that they laid out the goods. . .’)

e. as possessor
[awt m-nanaŋ-pampay-caw-na-ntut- ø] na
fire nr.dist-dur-appl-sit-dur-rm.past-i.sg poss
kalakn na-n-tay-mpi-yara-k
child.i.sg 3sg.nom-3sg.erg-see-seq-pick.up-irr
‘The son of him who was tending the fire found it.’
f. as possessed
manm p-ka-tay m-ɲa
cult.house.vii.sg vii.sg.nom-1sg.erg-see nr.dist-poss
[m-mpu-tkam-r-m]
nr.dist-3pl.erg-show-perf-vii.sg
‘I saw the cult house, yours, which they showed.’
Figure 23.6 illustrates the proposed representation for the relative clause of
the second example in (28b).
What is unusual about Yimas in comparison to more familiar languages is
that the entire relative clause, with the exception of the ‘leaked’ core
argument narmaŋ ‘woman’, corresponds to a single word, and such words
can be quite complex (28a, e). But that is entirely in keeping with the
polysynthetic typological profile of the language; the word is its favoured
grammatical unit and structures above that are very weakly syntactically
developed. This fact should again caution us against building grammatical
theories solely on the basis of what is familiar to us, such as constituency:
they blind us to other possibilities.
The final type of relative clause to consider are those in which the relativ-
ized noun is a temporal expression. These are obliquely case-marked like
relative clauses on locative adjuncts, and, also like them, the head of these
temporal relative clauses is a fossilized noun class and number suffix, of the
form -mp vii.sg. But in this case we know the noun to which it refers
because it is still synchronically in the language: pucm ‘part, piece, time’.
All the semantic relations on the Interclausal Semantic Relations Hierarchy
that are expressed typically by adverbial subordinate clauses in more famil-
iar languages, circumstances, reasons, conditionals and hypotheticals and
concessives, are expressed by this type of temporal relative clause in Yimas.
The differences between them are captured in the choice of tense and mood
for both the relative clause and the matrix clause. For example, for temporal
circumstances, the verb of the relative clause occurs with one of the past or
present tenses or that of the definite future.
(29) [m-mpu-ŋa-na-tay-ɲc-mp-n]
nr.dist-3pl.erg-1sg.acc-pres-see-pres-vii.sg-obl
pu-ka-apan-kt
3pl.nom-1sg.acc-spear-rm.fut
‘When they see me, I will spear them.’
But conditionals and hypotheticals are marked with mood affixes that
mark unreal or as yet unrealized events.

RX
COREN
CLAUSE NUCLEUSN
CORE
NUCLEUS
PRED
RX V RX N RX
m- n- tpul-c-
t l-c-
tpu -(n)ak
-(n
(n)ak narma
narm
r
NR.DIST 3SG.ERG hit-PERF I.SG wom
woman
NUCLEUS NUCLEUSN
CORE COREN NUM
TENSE CLAUSE
EIC
DEIC RX
Figure 23.6 Constituent structure of a relative clause (28b) in Yimas
(30) [m-mpu-ya-kr-mp-n Wamur-mat]

nr.dist-3pl.erg-come-rm.fut-vii.sg-obl Wambramas-i.pl
mambayŋki wunt kantk
banana.vi.pl sago.grub.v.pl with
pay-pra-kt ama-na-ra
carry-toward-rm.fut 1sg-poss-v.sg
‘If the Wambramas people will come, they can bring my bananas and
sago grubs.’

Counterfactuals are slightly different. They require the potential modality

prefix ant-, which expresses that an event is unlikely to happen, to co-occur
with a past- or present-tense marker on the verbs of both the relative clause
and the matrix clause. As the potential modality marker is a prefix, it
triggers rearrangement on verbs of the matrix clause of the normally left-
most bound nominative pronominal prefix to right edge.
(31) [tuŋkuruŋ ant-ka-tay-c-mp-n]

eye.vi.sg pot-1sg.erg-see-perf-vii.sg-obl
ant-ka-tu-r-(n)ak
pot-1sg.erg-kill-perf-iii.sg.nom
‘If I had seen the eye (of the crocodile), I would have killed it.’
Finally, concessives are formed by having the polarity of the relative

clause being positive, and that of the matrix clause, negative.
(32) [kay i-ka-ak-r-mp-n]

canoe.viii.sg viii.sg.nom-1sg.erg-push-perf-vii.sg-obl
arm-n ta-ka-wul-c-i
water-obl neg-1SG.erg-put.down-perf-viii.sg
‘Although I pushed the canoe, I didn’t put it down into the water.’
23.8 Conclusion
Yimas is an intriguing language that has much to offer grammatical theory

in its unusual grammatical structures due to its complex and rare mix of a
polysynthetic verb structure with an elaborate noun-class concordial system.
Sadly, it is now rapidly approaching extinction, but fortunately it was
documented when it was still in a relatively vibrant state and its full
richness able to be appreciated. Without the input of data from languages
like Yimas, our grammatical theory will be very impoverished indeed. But
many such crucial languages like Yimas are dying before our very eyes.
Grammatical theory is a scientific endeavour worthy of pursuit, but unless
we document the full range of grammatical possibilities, it will never be
complete. We may never know the grammatical riches the human language
capacity is able to evolve. And time is rapidly running out to do that. Our
discipline needs seriously to look at its priorities.
References
Bresnan, Joan and Lioba Moshi. 1990. Object asymmetries in comparative

Bantu syntax. Linguistic Inquiry 21: 147–185.
Broadwell, George Aaron. 2006. A Choctaw Reference Grammar. Lincoln, NE:
University of Nebraska Press.
Dixon, Robert M. W. 1979. Ergativity. Language 55: 59–138.

Duranti, Alessandro and Ernest Byarushengo. 1977. On the notion of ‘direct

object’. In Ernest Byarushengo, Alessandro Duranti and Larry Hyman
(eds.), Haya Grammatical Structure, 54–71. Los Angeles: Southern California
Occasional Papers in Linguistics.
Foley, William. 1986. The Papuan Languages of New Guinea. Cambridge: Cam-
Foley, William. 1991. The Yimas Language of New Guinea. Stanford, CA: Stan-
ford University Press.
Foley, William. 2005. Linguistic prehistory in the Sepik Ramu basin. In
Andrew Pawley, Robert Attenborough, Jack Golson and Robin Hide (eds.),
Papuan Pasts: Cultural, Linguistic and Biological Histories of Papuan-Speaking
Peoples (Pacific Linguistics 572), 109–144. Canberra: The Australian
National University.
Foley, William. 2017a. The languages of the Sepik-Ramu basin and environs.
In Bill Palmer (ed.), The Languages and Linguistics of the New Guinea Area:
A Comprehensive Guide, 177–412. Berlin: de Gruyter.
Foley, William. 2017b. Yimas: The profile of a polysynthetic language of New
Guinea. In Nicholas Evans, Michael Fortescue and Marianne Mithun (eds.),
Handbook of Polysynthesis, 808–829. Oxford: Oxford University Press.
Foley, William and Robert D. Van Valin, Jr. 1984. Functional Syntax and
Hale, Kenneth. 1976. The adjoined relative clause in Australia. In Robert
Hale, Kenneth. 1983. Warlbiri and the grammar of non-configurational
languages. Natural Language and Linguistic Theory 1: 5–47.
Martin, Jack B. 2011. A Grammar of Creek (Muskogee). Lincoln, NE: University of
Nebraska Press.
Munro, Pamela. 1984. The syntactic status of object possessor raising in West-
ern Muskogean. Proceedings of the Berkeley Linguistics Society 10: 634–649.
Munro, Pamela and Lynn Gordon. 1982. Syntactic relations in Western
Muskogean: A typological perspective. Language 58: 81–115.
Pawley, Andrew. 1993. A language which defies description by ordinary
means. In William Foley (ed.), The Role of Theory in Language Description,
87–130. Berlin: de Gruyter.
In Balthasar Bickel, Lenore Grenoble, David Peterson and Alan Timberlake
(eds.), Language Typology and Historical Contingency, 91–124. Amsterdam:
John Benjamins.

24
of Avatime (Kwa,
Niger-Congo, Ghana)
Saskia van Putten and Rebecca Defina
Abbreviations

ing-rules.php) with the following additions:
C noun class IT itive

CLM clause linkage marker REC recurrent
CM clause marker SVM serial verb marker
HAB habitual VEN ventive
INT intentive
24.1 Introduction
In this chapter we give an overview of the grammar of Avatime from a Role

and Reference Grammar (RRG) perspective. Avatime belongs to the Kwa
branch of the Niger-Congo language family. Within Kwa, it belongs to the
Ka-branch of the group of Ghana-Togo Mountain (GTM) languages. It is
spoken in the south-east of Ghana, by about 15,000 speakers.
As a background to the discussion of the grammar, two properties of
Avatime phonology are important to mention. First, Avatime is a tone
language. It has three level tones: low (marked à), high (unmarked) and
extra-high (marked á). The extra-high tone has a limited distribution and
occurs mostly on grammatical morphemes and function words. Second,
there is advanced tongue root (ATR)-based vowel harmony. Vowels in most
affixes and clitics harmonize in their ATR value with the nearest root vowel.
þATR vowels are i, e, o, u and ATR vowels are i, ɛ, a, , u. c
_ _
The information presented in this chapter is based on data collected
during several field trips to the Avatime area between 2008 and 2013.1 More
elaborate descriptions of the grammar of Avatime can be found in Defina
(2016a) and Van Putten (2014a).

890 SASKIA VAN PUTTEN AND REBECCA DEFINA
Part I of this chapter gives an overview of the basic grammatical structures

of simple sentences. It describes lexical categories, predicate classes and
logical structure, the layered structure of the clause, including the operator
projection, referential phrases and adpositional phrases and the linking
from semantics to syntax in simple sentences. Finally, it examines focus
structure. Part II discusses the syntax and semantics of serial verb construc-
tions and the questions they raise for the Interclausal Relations Hierarchy
(Van Valin 2005: 209).
PART I. BASIC GRAMMATICAL STRUCTURES
24.2 Lexical Categories
Avatime has the distinct lexical categories of noun, verb, adjective,

ideophone, adverb, preposition and postposition. Nouns are morphologic-
ally distinguished by their noun-class prefixes (see Section 24.5.1) and can be
modified by adjectives, numerals and determiners. Nouns are used as refer-
ring expressions. They cannot function as predicates or modifiers and there
is no way to derive other lexical categories from nouns.
Verbs are distinguished from other lexical categories by their ability to be
marked with a subject prefix and aspect, modality, negation and direction
markers (see Section 24.4.2). Verbs are used as predicates. They can be used as
referring expressions only when they are nominalized. To fully nominalize a
verb, the verb is reduplicated and optionally marked with the noun-class
prefix ku-/ku-. There are two constructions where the verb is marked with a
_
noun-class prefix without being reduplicated: in non-finite complement
constructions, where it is marked with the prefix ku-/ku- and in the focus
_
construction (see Section 24.7), where it is marked with the prefix ki-/ki-. In
_
these cases, the verb appears to be nominalized to a certain extent, but it
cannot function as a full noun in this form.
Avatime has a small class of non-derived and non-ideophonic adjectives.
These modify nouns and cannot be used by themselves as predicates or
referring expressions. They do not share any morphological properties to
distinguish them from other lexical categories. The majority of property
words are not adjectives, but verbs or ideophones. Verbs which include a
property or a state as part of their meaning can be turned into adjectives by
reduplication, such as kpakpa ‘dry’ from the verb kpa ‘to dry’. Adjectives can
be nominalized by prefixing them with a noun-class prefix. For instance, the
̣
adjective sı̀sàmi ̣
‘small’ can be turned into -sı̀sàmi ‘small one’ (referring to
c
something of class 1 singular).
Avatime also has a small class of non-ideophonic adverbs. Like the adjec-
tives, they do not share any morphological properties, but they are distin-
guished by their function as modifiers of predicates or clauses. These
adverbs cannot be used as predicates, referring expressions or modifiers of
nouns. Some examples are nyafɛ ‘maybe’, àbla ‘now’ and kóko ‘already’.

A Grammatical Sketch of Avatime 891
There is a large class of ideophones that overlaps in function with the

classes of adjectives and adverbs. Ideophones are ‘marked words that depict
sensory imagery’ (Dingemanse 2012, see Chapter 9 of this volume). They can
be used to modify nouns, predicates and clauses, but they cannot be used as
predicates or referring expressions. Ideophones are distinguished from
other word classes by their marked phonology (e.g. long words, repeated
syllables, long final vowels). Some examples are pititi ‘white’, ha~a~a~a~
___
‘intensely’ and trátrátrátrá ‘very neat’.
Finally, Avatime has small sets of prepositions and postpositions. There
are only two prepositions: locative ní and comitative (a)nı̀. There are five
postpositions, four of which have evolved from nouns. An example is ese
‘under’, which comes from the noun kese ‘ground’ (for examples of adposi-
tions see Section 24.5.2). As we show in Section 24.5.2, the postpositions are
very noun-like in their use and can be analysed as heading referential
phrases (RPs). Nevertheless, they are a separate lexical category on both
morphological and syntactic grounds. Morphologically, they are distinct
from nouns as they have lost their noun-class prefixes. Syntactically, they
cannot be modified with nominal modifiers such as definite articles and can
only occur in the position following another referential phrase.
24.3 Predicate Classes and Logical Structure
We distinguish four main Aktionsart classes in Avatime: states, activities,

semelfactives and accomplishments. The primary distinguishing factor is
their interpretation with progressive and perfective aspects. For more detail
about Avatime Aktionsart classes, see Defina (2009, 2018).
States are interpreted as present in the perfective, as can be seen in the
example in (1). They can be marked as progressive, where they are inter-
preted as persistent states (e.g. ‘the building is still in Vane’).
(1) ligba¼lὲ li-lí ̣ ní ̀ ʋan ̀

c c
_ _
c3.sg-building¼def c3.sg.sbj.pfv-be.at loc Vane
‘The building is in Vane.’
Logical Structure: be.at′ ( ̀ ʋan ̀ , ligbalὲ)
c c
_
Activities are dynamic events without an inherent endpoint. They are
generally interpreted as occurring in the past in the perfective, as in
example (2), and as occurring in the present in the progressive (e.g. ‘Afua
is eating rice’).
(2) àfua a-tá ki-mimi

_ _ _
Afua c1.sg.sbj.pfv-chew c4.sg-rice
‘Afua ate rice.’
Logical Structure: do′ (Afua, [eat′ (Afua, kimimi)])
_ _ _
Accomplishments are state-change events with internal duration. Like
states, they are typically interpreted as present in the perfective, as can be

seen in example (3), but they are interpreted like activities in progress with
the progressive (e.g. ‘the door is opening’).
(3) ò-pópo¼lò ɛ-drà

c2.sg-door¼def c2.sg.sbj.pfv-open
‘The door is open.’
Logical Structure: BECOME open′ (òpópo)
Semelfactives are events that lack duration. Like activities, they are inter-
preted as past actions in the perfective (e.g. ‘she came’). However, unlike
activities, they cannot be modified by duration adverbials such as ‘for 5
minutes’ and ‘quickly’. They can be used with the progressive, in which case
they are interpreted as iterative or repeated actions, as in example (4).
(4) o-ne¼e ὲέ-ba

c2.sg-mother¼def c1.sg.sbj.prog-come
‘The mother comes regularly.’
Logical Structure: SEML come′ (one)
As yet, no clear diagnostic for the classes achievement and active accom-
plishment has been found for Avatime. The typical test using the modifiers
‘for X time’ or ‘in X time’ is not applicable as this specific semantic distinc-
tion is not consistently made. There is, however, a group of atomic telic
predicates where the endpoint is reached in a single step rather than
gradually (Caudal and Nicolas 2005), which can be distinguished based on
their incompatibility with the adverb petee ‘completely’, as in example (5).
(5) osófo¼e a-trɛ ní òholò (*petee)

pastor¼def c1.sg.sbj.pfv-go loc Ho (*completely)
‘The pastor went to Ho (*completely).’
Logical Structure: do′ (osófo, [go′ (osófo)]) & INGR be.at′ (Ho, osófo)
In addition to the Aktionsart predicate classes discussed above, it is worth

mentioning causative predicates. Some verbs, such as tr ‘put on’ and bàsi
c
_
‘show’ are always causative. Other verbs have a non-causative reading when
they are used intransitively and a causative reading when they are used
transitively, such as dra ‘open’ in example (6) (see also (3) above).
(6) -d ̀ bàsi¼ɛ
c c a-dra ò-pópo¼lò
_
c1.sg-teacher¼def c1.sg.sbj.pfv-open c2.sg-door¼def
‘The teacher opened the door.’
Logical Structure: do′ ( ̀ d ̀ bàsi, Ø) CAUSE [BECOME open′(òpópo)]
c c
24.4 Clause Structure and Operators
24.4.1 Clause Structure

Canonical constituent order in Avatime is SVO: the privileged syntactic
argument (‘subject’; PSA) precedes the verb, and other arguments

Sentence
Clause
RP Core Periphery
Pro Nuc RP
Pred
V PP
ɔd
́ zɛɛ á- ta áʋanà ní lịgbalɛ mɛ̀
Figure 24.1 The layered structure of the sentence in example (7)
(‘direct/indirect object’) follow the verb. Adverbial phrases follow the post-
verbal arguments. There is head-marking of the PSA on the verb. A canonical
transitive sentence with an adverbial phrase in Avatime can be seen in
example (7) and the syntactic structure of this example is shown in
Figure 24.1. Following Van Valin (1985) and Van Valin and LaPolla (1997),
we analyse the PSA-indexing prefix on the verb as an argument in the core,
whereas the RP with which this prefix is coreferential is analysed as occur-
ring outside the core, but inside the clause. This captures the observation
that a lexical RP or independent pronoun is not necessary, and example (7)
remains grammatical when leaving out the RP ́ dzɛɛ ‘the woman’. c
́ -dzɛ¼ɛ á-ta á-ʋa¼nà ní li-gba¼le

c
(7)
_
c1.sg-woman¼def c1.sg.sbj.pfv-chew c3.pl-bean¼def loc c3.sg-room¼def
mὲ
inside
‘The woman ate beans in the room.’
Avatime makes use of the pre-core slot (PrCS) for question formation and
focus marking (see also Section 24.7 and Van Putten 2016). The question
word or focused RP is placed in the pre-core slot and marked with a final
extra-high tone, as shown in examples (8) and (9).
egé ́ -dzɛ¼ɛ a-ta
c
(8)
what c1.sg-woman¼def c1.sg.sbj.pfv-chew
‘What did the woman eat?’ (elic-QUIS-foc_100714_SO)
á-ʋa¼ná ́ dzɛ¼ɛ a-ta

c
(9)
c3.pl-bean¼def:foc c1.sg-woman¼def c1.sg.sbj.pfv-chew
‘The woman ate [the beans]FOC.’ (elic-QUIS-foc_100714_SO)
The pre-detached position is frequently used to host topical expressions or

expressions encoding spatial or temporal background information (see Van
Putten 2014b). When an argument that is not the PSA is pre-detached, a

Sentence
PrDP Clause
PrCS RP Core
Pro Nuc
Pred
RP RP V
kivoe áʋaná ɔd
́ zɛɛ á- ta
Figure 24.2 A sentence with pre-detached position and pre-core slot (example 11)
resumptive pronoun occurs in its regular postverbal position, as shown in

example (10). Example (11) shows a sentence with both focus-marking and a
pre-detached element. The syntactic structure of this example is shown in
Figure 24.2. Note that the PrCS is not the same structural position as the
regular position for PSA RPs, in that it is also within the clause but outside
the core. Focused elements always precede the PSA RP.
á-ʋa¼nà, ́ -dzɛ¼ɛ a-ta na

c
(10)
c3.pl-bean¼def c1.sg-woman¼def c1.sg.sbj.pfv-chew c3.pl
‘The beans, the woman ate them.’
kivoe á-ʋa¼ná ́ -dzɛ¼ɛ á-ta

c
(11)
yesterday c3.pl-bean¼def c1.sg-woman¼def c1.sg.sbj.pfv-chew
‘Yesterday, the woman ate [the beans]FOC.’
24.4.2 Operators
In this section, we give an overview of the operators that have scope over the
different layers of the layered structure of the clause. The Avatime verb is
usually marked with a prefix that indexes the PSA. This prefix (henceforth
referred to as the PSA prefix) also encodes aspect or mood. It is optionally
followed by other aspect, mood and directionality prefixes. There is no tense
marking. An overview of the categories marked on the verb and the order in
which they occur can be seen in (12) and an example with all slots occupied
can be seen in (13).
(12) PSA þ Perfective/Progressive/Habitual/Potential/Subjunctive þ (Negation) –

(Intentive) – (Recurrent) – (Directional) – Root
(13) m ́ -tá-zɛ̌ -zɛ-panı̀ ̣

c w c
1.sg.sbj.pfv.neg-int-rec-it-talk 2.sg
‘I will not be going to talk with you.’ (R0811291)

Table 24.1 Paradigms of PSA prefixes (most noun classes omitted)
Set 2
- Negative (with extra-high Set 3
tone) - Subjunctive
Set 1 - Subjunctive (some verbs) (some verbs)
Person/ number/ Perfective - Perfective (locative and - Habitual (with
noun class (most verbs) copula verbs) zɛ̌ - prefix) Potential Progressive
1s me-/ma- mo-/m - c mi-/mi- máà- mèé-/mὲέ-

1p ki-/ki ku-/ku- ki-/ki-_ ̣
kíà- kìí-/kìí-̣ ̣
_ _
2s wo-/w -c wo-/w_ -c wu-/wu- wáà- wèé-/wὲέ-
2p mle-/mlɛ- mla mli-/ml_i- mláà- mlèé-/mlὲέ-
C1.SG e-/a- o-/ -
c i-/i- _ áà- èé-/ὲέ-
C1.PL be-/bɛ- ba _ i-
bi-/b bíạ̀ bèé-/bὲέ-
_
...
There are five paradigms of PSA prefixes, as shown in Table 24.1. For the
sake of brevity, the forms for the noun classes other than class 1 singular
and plural have been omitted (a full table can be seen in Van Putten 2014a or
Defina 2016a). As the table shows, there is no one-to-one correspondence
between each paradigm and an aspect/mood category. It is also clear that the
prefixes are formally related to each other; it is likely that they have evolved
from combinations of a prefix indexing the PSA plus an aspect/mood
marker, but they cannot be synchronically analysed as such.
We will now briefly discuss the operators one by one. For more detail
about these markers and their semantics, see Defina (2018).
Given that aspect marking modifies the internal temporal structure of the
event itself, the aspect markers are considered to be nuclear-level operators
(Van Valin and LaPolla 1997; Van Valin 2005). Avatime has four aspectual
categories: perfective, progressive, habitual and recurrent. Their forms are
described in (14).
(14) Aspect marking

• The perfective is marked with a PSA prefix from set 1 or, in the case of
locative and copula verbs, set 2. It is the most frequently occurring verb
form in Avatime, some examples of which have already been presented
in (7)–(11).
• The progressive has its own paradigm of PSA prefixes, as shown in
Table 24.1.
• The habitual uses a PSA prefix from set 3, combined with the prefix zě-/zɛ̌ -,
see example (15).
• The recurrent is one of the two optional aspect/mood prefixes and indicates
repeated or recurring actions and situations. It is marked with the prefix
zě-/zɛ̌ -,2 which is always combined with one of the obligatory aspect/mood
categories, see example (16).
(15) mí-zě-do mawùye li-bo¼lè ní s ̀ lὲ mὲ c

1sg.sbj.hab-hab-say God c3.sg-word¼def loc church inside
‘I preach the word of God in church.’ (life_100614_WE)

(16) ma-zɛ̌ -wà a-xwὲ¼na ní cocoa marketing board

1.sg.sbj.pfv-rec-do c3.pl-work¼def loc cocoa marketing board
‘I used to work at the cocoa marketing board.’ (life_AB)
There are two directionals: the itive (‘away’), marked with the prefix ze-/zɛ-
and the ventive (‘towards’), marked with the prefix bá-/bé-, shown in example
(17). These forms express orientation or motion away from or towards the
deictic centre. This means they are core-level operators (see Van Valin and
LaPolla 1997: 45).
(17) a-bá-feke télefòn¼ye

c1.sg.sbj.pfv-ven-pick.up telephone¼def
‘He came and picked up the telephone.’ (finsto_100716_DQ)
Another core operator is the intentive, which indicates the intention of

the actor to carry out the action expressed by the verb. The intentive is
marked with the optional prefix tá- which combines with either the
perfective aspect or the subjunctive mood. For an example of the former,
see (13).
Negation is expressed by a PSA prefix from set 2 with an extra-high tone.
An example can be seen in (13). When the progressive is negated, this PSA
prefix is followed by the prefix li-/li-. When the habitual is negated, the prefix
_
zě-/zɛ̌ - is deleted, leaving only its tone and lengthening the vowel of the PSA
prefix. Negation marking can be used for both core-level and clause-
level negation.
Avatime has three clause-level operators marked on the verb: potential,
subjunctive and imperative. Their properties are described in (18).
(18) Clause-level operators marked on the verb

• The potential has its own PSA agreement paradigm as shown in
Table 24.1. The potential PSA prefix cannot be combined with negation
marking. For a negative potential interpretation, the negative perfective
intentive is used.
• The subjunctive is marked with a subject prefix from set 2 for some verbs
and for other verbs with a subject prefix from set 3. An example is shown in
(19). When the subjunctive is negated, a prefix ku-/ku- is added while the PSA
_
prefix remains the same. The subjunctive has irrealis and illocutionary
force interpretations.
• The imperative is marked by the absence of a PSA prefix, as in ba ‘Come!’. It
is only used for second-person singular commands. For other commands,
the subjunctive is used. To negate imperative clauses, the negative
subjunctive is used.
(19) mla-trɛ
2.pl.sbj.sbjv-go
‘You (pl) should go.’ / ‘Go! (to multiple people)’
The final operator to be discussed here is question marking, an illocu-

tionary force operator. Polar questions are marked with a clause-final

particle na or, more frequently, with intonation only. Content questions are
marked with a question word in the pre-core slot (see example (8) in Section
24.4.1).
According to RRG (Van Valin and LaPolla 1997; Van Valin 2005), the
order of operators in a given language always follows the same pattern
whereby clausal operators occur furthest from the verb, core operators
closer than clausal operators, and nuclear operators are the closest to the
verb. In Avatime, this prediction is mostly borne out, but there are two
exceptions: (i) the (core) directional markers occur closer to the verb than
the (nuclear) aspect markers, and (ii) the (core) intentive marker also
occurs closer to the verb than the (nuclear) perfective marker. The direc-
tional markers have clearly grammaticalized from motion verbs (ba ‘come’
and za ‘pass’). Before grammaticalization, they would have occurred as the
first verb in a serial verb construction (SVC), directly followed by the
second verb (see Sections 24.9 and 24.10 for more information about serial
verb constructions). As all aspect and mood marking in SVCs is on the first
verb, it now precedes these directional prefixes. A similar explanation
could account for the intentive following the perfective: the intentive tá-
is also likely to be the result of a grammaticalization process from a serial
verb construction, probably from the verb trɛ ‘go’ or tráà ‘come’. Another
possible explanation for the intentive following the perfective is the
hypothesis that the perfective forms used to be the bare subject agreement
markers unmarked for aspect/mood, as is the case in many related lan-
guages (Welmers 1973). Through language change, the default PSA prefix
came to be used solely for perfective marking and thereby turned into a
nuclear-level operator, which happened to occur in a position preceding
the intentive marker.
Figure 24.3 shows the sentence in example (13), repeated here as (20), with
both constituent and operator projections.
(20) m ́ -tá-zɛ̌ -zɛ-panı̀ ̣

c w c
1.sg.sbj.pfv.neg-int-rec-it-talk 2.sg
‘I will not be going to talk with you.’ (R0811291)
24.5 Referential and Adpositional Phrases
24.5.1 Referential Phrases Headed by Nouns

Like clauses, referential phrases (RPs) have a layered structure. The nucleus
of the Avatime RP can be either a noun or a postposition. When it is headed
by a postposition it functions as an adpositional phrase, and this type of RP
will be discussed in Section 24.5.2. This section will instead discuss RPs
headed by nouns.
Avatime nouns, like those in many Niger-Congo languages, take part in a
noun-class system. Each noun (with the exception of some loanwords)

Sentence
Clause
Core
NP Nuc NP
Pred
V
mɔ-́ tá-zɛ-̌zɛ- panì wɔ
V
ASP
Nuc
ASP Nuc
DIR Core
MOD Core
STAT Clause
Figure 24.3 Sentence from example (20) with constituent and operator projections
carries a noun-class prefix which indicates its gender and number. Within
the referential phrase, there is agreement on numerals, articles and demon-
stratives (see example (24)). For more detailed information on the noun-class
system and the various nominal modifiers, see Schuh (1995), Van Putten
(2014a) and Defina (2016a).
The constituent order in Avatime RPs is shown in (21). Adjectives are
considered to occur in a periphery of the nucleus, following Van Valin
(2005), whereas the other nominal modifiers are operators on different
layers of the RP.
(21) noun – adjectives – numeral – determiner – particles
The only operator on the nuclear level is number, which is marked on the
noun class prefix (e.g. ̀ -kli ‘leg’ vs. ı̀ ̣-kli ‘legs’). The numerals are operators on
c
_ _
the core level, as they provide quantification. The determiners are oper-
ators on the RP level, as they ground the RP in discourse. The class of
determiners contains the elements shown in (22), of which only one can
occur in a given RP.
(22) Determiners
• The definite article, which is a monosyllabic enclitic and agrees with the
noun class of the head noun.

• The indefinite article t , prefixed with a noun-class agreement prefix. This

c
article has a specific indefinite interpretation (e.g. ‘some’, ‘a certain’). For a
non-specific indefinite interpretation, the determiner slot is unfilled.
• The proximal demonstrative yà and distal demonstrative l ̀ , both prefixed
c
with a noun-class agreement prefix.
The class of particles contains words such as tsyɛ ‘also’, k ‘by contrast’ and c
kò ‘only’. Like definiteness and deixis markers, they ground the RP in
discourse and are therefore considered to be RP-level operators.
Possession in Avatime is mostly indicated by juxtaposition with a possessor–
possessed word order. The regular independent pronouns are used as posses-
sor pronouns. For possessed kinship terms, there is a slightly more complex
construction in which the possessor pronoun fuses with the noun-class prefix
of the possessed kinship term.3 With a nominal possessor, this fused possessor
pronoun is also present, and, thus, in these cases there is head-marking of the
possessor on the possessed noun. Examples of kinship and non-kinship terms
with nominal and pronominal possessors can be seen in (23).
(23) yɛ ‘c1.sg’ þ c
̀ -mà-n ‘the town’
c ! yɛ ̀ -mà-n ‘her town’
c c
yɛ ‘c1.sg’ þ o-ne ‘mother’ ! ye-ne ‘her mother’
c-nùv ̀ -ɛ ‘the child’
c þ li-kuto-lɛ ‘the hat’ ! nùv ̀ -ɛ li-kuto-lɛ ‘the child’s hat’
c c
c-nùv ̀ -ɛ ‘the child’
c þ o-ne ‘mother’ ! nùv ̀ -ɛ ye-ne ‘the child’s mother’
c c
The possessor occurs in the RP initial position (RPIP), analogous to both the
pre-detached position and pre-core slot in the layered structure of the clause
(see Van Valin 2005: 26).
An example of a complex RP in Avatime can be seen in (24). The constitu-
ent and operator projections of this example are shown in Figure 24.4.
(24) bá-dzɛ kpekpe tia-bà¼a ke-pe¼à

c1.pl.woman short c1.pl-two¼def c6.sg-house¼def
‘The house of the two short women.’
24.5.2 Adpositional Phrases

As mentioned in Section 24.2, there are both prepositions and postpositions
in Avatime. Both of Avatime’s prepositions – the comitative preposition (a)nı̀
‘and/with’ and the general locative preposition ní – are predicative prepos-
itions. This means they occur in the nucleus of a prepositional phrase (see
Figure 24.5 and Chapter 10 of this volume). The five locative postpositions
are indicators of what Ameka (1995) calls the search domain, that is, the part
of the reference object where something is located. Postpositional phrases
are not locative by themselves but behave like noun phrases; they can, for
instance, serve as arguments of a verb. Because of this, we analyse the
postpositional phrase as an RP with the postposition as its nucleus and the
noun expressing the location in an RP-initial position, analogous to posses-
sive constructions. This makes sense from a historical perspective, as most
postpositions have evolved from nouns which would have combined

RP
RPIP CoreR
RP
CoreR
NucR R NucR
N MP N
bá-dzɛ kpekpe tịabà =a ke-pe=à
N N
NUM NucR NUM NucR
CoreR QUANT CoreR
RP DEF RP DEF
Figure 24.4 Complex RP with constituent and operator projections (example 24)
PP
CoreP
RP
RPIP CoreR
NucP RP NucR
Pred
PREP POSTP
ní ɔk̀plɔnɔ ̀ abà
Figure 24.5 Constituent projection of a PP (example 25)
with the locatum in a possessive construction (e.g. ‘the upper surface of the
table’ ! ‘on the table’, see also Section 24.2).
In locative phrases, a postpositional phrase must be combined with the
preposition ní, as shown in example (25). The constituent projection of this

example is shown in Figure 24.5. Instead of a postposition, a noun can also

be used to indicate the search domain, as in example (26). This follows from
the analysis presented above: in this case the two nouns occur in a possessive
construction, as the situation with the prepositions would have been before
grammaticalization was complete. Thus, example (26) literally translates as
‘located at the back of the car’.
ní ̀ -kpl ¼n ̀ abà
c c c
(25)
loc c2.sg-table¼def on
‘on the table’
(26) ní ò-hui¼lò ke-de¼à

loc c2.sg-car¼def c6.sg-back¼def
‘behind the car’
24.6 Macroroles, Arguments and Linking from

Semantics to Syntax
In this section, we discuss how logical structure is linked to syntactic

structure, including how semantic macroroles are assigned in Avatime
and how the privileged syntactic argument is selected.
Macrorole selection in Avatime simple sentences follows the Actor-
Undergoer Hierarchy (AUH) shown in (27). The highest-ranking argument
is selected as the actor. When it comes to the selection of the undergoer, Van
Valin (2005) shows that, across languages, there are two possible principles:
either the lowest-ranking or the second-lowest-ranking argument is selected
as the undergoer. When a verb has only two core arguments, the two
principles will always select the same undergoer. However, a difference
emerges with three-place predicates. In Avatime three-place predicates, the
two non-actor arguments always follow the verb with the ‘indirect object’
preceding the ‘direct object’, as shown in example (28). The logical structure
of this sentence is shown in (29) and makes clear that the highest-ranking
argument is Kofi, the second-highest is the first-person singular pronoun and
the lowest-ranking argument is egumena ‘the cows’. Given that there is no
case marking and no indexing of either of the two arguments on the verb,
there doesn’t seem to be clear evidence for which of the two is the under-
goer and therefore which of the two principles applies.
(27) The Actor-Undergoer Hierarchy (based on Van Valin and LaPolla 1997 Fig. 4.2)
ACTOR UNDERGOER
----------------------------------------------------->
<- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Arg. of 1st arg. of 1st arg. of 2nd arg. of Arg. of state

DO do′ (x,. . .) pred′ (x, y) pred′ (x, y) pred′ (x)
[‘------------>’ ¼ increasing markedness of realization of argument as macrorole]

(28) kòfi a-bàsi mɛ e-gume¼nà

_
Kofi c1.sg.sbj.pfv-show 1.sg c3.pl-cow¼def
‘Kofi showed me the cows.’
(29) do′ (Kofi, Ø) CAUSE [see′ (1s, egumenà)]
After assigning macrorole status, the PSA is selected. In Avatime, the PSA is
the argument that is indexed on the verb. Avatime has a nominative-
accusative alignment pattern: if there is only one macrorole argument, this
is selected as the PSA, and if there are two macroroles, the actor is selected as
the PSA. The PSA is invariable: there is no passive construction in Avatime
and non-macrorole PSAs are not allowed. In example (28), the actor, Kofi, is
selected as the PSA.
After the selection of the PSA, the appropriate syntactic template is
selected. The syntactic template for a canonical Avatime sentence has one
core slot before the verb for a head-marked argument and up to two core
slots for RP arguments after the verb (see Section 24.4.1). In the case of
example (28), a syntactic template with two core slots after the verb is
chosen, as there are three arguments in the logical structure. The preverbal
head-marking slot is linked to the PSA, which is Kofi. This slot can only
contain PSA-indexing prefixes and does not allow referential phrases. There-
fore, it is filled with a PSA-indexing prefix of noun class 1 singular (class 1 is
used for people) and the RP Kofi is placed in the pre-core position (see also
Section 24.4.1). The undergoer and non-macrorole arguments are linked to
the two postverbal core-internal slots.
24.7 Focus Structure
In addition to the constituent projection and operator projection, RRG

posits a third component of grammar, representing the information struc-
ture of the sentence: the speech-act projection (see Chapter 11 of this
volume). In this section, we briefly discuss focus marking in Avatime. For
more detail, see Van Putten (2016).
The potential focus domain in Avatime canonical clauses coincides with
the entire clause, that is, any element of a clause can potentially be in focus.
Focused elements in canonical clauses are not marked as such with inton-
ation; the focus interpretation comes purely from the context.
To make clearer which part of the sentence is in focus, a syntactic focus
construction can be used (see also Section 24.4.1). This construction is often
used when the focus is contrastive. In this construction, a constituent is
placed in the pre-core slot and marked with a final extra-high tone. This
structure is most frequently used for narrow focus, that is focus on the
constituent in the pre-core slot only, as was shown in example (9) in
Section 24.4.1. To mark narrow focus on the verb, a copy of the verb root,
prefixed with the noun class prefix ki-/ki- is placed in the pre-core slot and
_

marked with the final extra-high tone, while the inflected verb remains in
its normal position. An example is shown in (30), where the focus is on the
lexical content of the verb, but the same construction can also be used
to mark focus on an aspect, modality or status operator (see also Van
Putten 2016).
(30) ki-h ́
c bɛ-tá-h c lc
c4.sg-grind:foc c1.pl.sbj.pfv-int-grind c3.sg
‘Will they [grind]FOC it (or pound it)?’ (illness_100616_SO-DS)
The focus construction may also be used for sentence focus or predicate
focus (focus on the verb þ argument that is not the PSA or PP). A sentence
focus interpretation is possible when the PSA is marked for focus and a
predicate focus interpretation is possible when an argument that is not the
PSA or a peripheral PP is marked for focus. An example of the latter can be
seen in (31).
(31) (One speaker is telling a story in which a boy riding a bicycle meets a girl. The
listener asks a clarification question.)
yɛ tsyɛ, gas ́ c ὲέ-kpɛ
c1.sg too bicycle:foc c1.sg.sbj.prog-put
‘Is she also [riding a bicycle]FOC ?’ (pear_100630_GoD-FB)
For the linking from semantics to syntax, the use of the focus construction
means that a syntactic template with a pre-core slot will be chosen (see also
Section 24.4.1). If the element to be placed in the pre-core slot is a non-PSA
argument, the number of postverbal core slots will be reduced by one. PSA
indexing on the verb is obligatory, so the preverbal core slot which contains
the verb prefix cross-referencing the nominal PSA argument is always filled,
irrespective of whether the PSA occurs in the pre-core slot or in its regular
pre-core position.
24.8 Summary
In Part I of this grammatical sketch, we have discussed the grammar of

Avatime simple clauses. Avatime is a language with a clear distinction
between nouns and verbs, small lexical categories of adjectives, adverbs
and adpositions and a large category of ideophones. It has a rather rigid
constituent order, in which the PSA precedes the verb and other arguments
follow the verb, followed by adverbial phrases. The PSA is indexed on the
verb with head-marking. Question words and optionally focused elements
occur in the pre-core slot.
Referential phrases can have either a noun or a postposition as their
nucleus. If a noun heads the RP, it can be followed by a number of modifiers,
most of which agree in noun class with the head noun. Possession is
indicated by juxtaposition with possessor–possessed word order. When a
postposition heads the RP, no modifiers are possible and the postposition is

always preceded by another RP that functions like a possessor. Locative

phrases are formed with the locative preposition ní.
Most operators of the layered structure of the clause are marked on the
verb by the choice of PSA prefix out of several possible paradigms and/or a
specialized prefix. The order of operators does not always follow the
expected clausal-core-nuclear order. This may be due to the origin of some
prefixes as verbs in serial verb constructions and to the change from an
unmarked to a marked perfective.
PART II. THE SYNTAX AND SEMANTICS OF SERIAL

VERB CONSTRUCTIONS
In this part, we discuss the relation between syntax and semantics in serial
verb constructions (SVCs). First, we describe the properties of SVCs in Ava-
time and how they are distinguished into syntactic subtypes (Section 24.9).
We then discuss the semantic functions of Avatime SVCs and how they fit
within the Interclausal Relations Hierarchy connecting syntactic structures
and semantic functions (Section 24.10).
24.9 Serial Verb Constructions
Serial verb constructions are sequences of two or more finite verbs in one
clause without overt marking for coordination or subordination. In Ava-
time, only the first verb is fully inflected. Subsequent verbs are either bare,
as in example (32) or marked with a reduced agreement prefix, as in
example (33). These reduced prefixes are specific to SVCs (for more infor-
mation, see Defina 2016b). All verbs share their PSA and may also share
another argument, as is the case with example (33).
(32) o-di ŋwὲ

c1.sg.sbj.pfv-sit drink
‘S/he sits drinking.’
(33) a-ŋwya ki-dít ca-kpέ

c1.sg.sbj.pfv-throw c4.sg-thing:indf svm.c1.sg.pfv-put.in:loc
c-kà¼ɛ
c1.sg-father¼def
‘She threw something to the father.’ (contrexp-28_120912)
Avatime SVCs can be divided into three subtypes: nuclear, core and
sequential, based on a number of morphosyntactic differences. In what
follows we briefly describe the properties of the different types of SVC and
show how they can be analysed in RRG. For a more elaborate account, see
Defina (2016b).
The different subtypes can be distinguished by examining their behav-
iour with operators at different levels. Negation and most aspect and mood

markers are fused to some degree with the PSA agreement markers (see
Section 24.4.2) and can only be marked on the first verb of an SVC. These
markers always have broad scope over the whole SVC. Even the intentive,
which is the only morphologically separate mood marker, can only occur
with the first verb and scopes over the whole construction. This suggests
that the verbs in all Avatime SVCs are linked below the clause level.
Aspect and aspectual adverbials are nuclear-level operators and can be
used to distinguish a set of nuclear-level SVCs from the rest. Many SVCs
allow aspectual adverbials between the verbs with narrow scope over a
single verb, as can be seen in example (34). The recurrent aspect can also
be used to mark each verb in these SVCs individually, as can be seen in
example (35). Neither of these is possible with the nuclear-level SVCs. While
aspectual adverbials are occasionally accepted between the verbs of these
nuclear SVCs, they always have broad scope over the whole construction and
the recurrent aspect can only be marked on the first verb, as can be seen in
examples (36) and (37). This indicates that these SVCs are formed through
nuclear co-subordination while the other Avatime SVCs are joined at a
higher level.
(34) e-bu àgbèlı̀¼ye ı̀-dru¼lè kóko kí ̣ Kwami

c1.sg.pfv-remove cassava¼def c2.pl-mound¼def already give Kwami
‘He already dug cassava mounds for Kwami.’ (But may keep them for himself )
(35) a. mà-zɛ̌ -dzɛ Òholò a-wà à-xwὲ¼na

1.sg.pfv- REC -go Ho svm.1.sg.pfv-work c3.pl-job¼def
‘I was going (repeatedly) to Ho and working.’
b. mà-dzɛ Òholò a-zɛ̌ -wà à-xwὲ¼na
1.sg.pfv-go Ho svm.1.sg.pfv- REC -work c3.pl-job¼def
‘I went to Ho and was working.’ (moved to Ho for some time)
(36) ba-dí ̣ koko gu ku-nugu¼yò

c1.pl.pfv-sit already talk c5.sg-mouth¼def
‘They already sat talking.’
(37) a. ba-zɛ̌ -dí ̣ ŋwὲ kù-gòda

c1.pl.pfv- REC -sit drink c6.sg-palmwine
‘They were sitting drinking palmwine.’
b. *ba-dí ̣ zɛ̌ -ŋwὲ kù-gòda
c1.pl.pfv-sit REC -drink c6.sg-palmwine
Directional prefixes and locational and temporal adverbials are core-

level operators and their scope properties distinguish the final two sub-
types. Sequential SVCs can occur with distinct locational and temporal
adverbials modifying each verb phrase, as can be seen in example (38).
These SVCs also allow the directional prefix to occur on the second verb
with narrow scope, as can be seen in example (39). This suggests these SVCs
consist of distinct cores with their own peripheries and so we analyse them
as core coordinations.

(38) ma-tsà tomatoes¼ye ní li-vlɛ¼lὲ

_
1.sg.sbj.pfv-cut tomatoes¼def loc c3.sg-morning¼def
à-kpɛ ní kè-zi¼a mὲ áblà
svm.1.sg.pfv-put loc c6.sg-bowl¼def inside now
‘I cut the tomatoes in the morning and put them in the bowl now.’
(39) ma-k ̀c kà-wɛ¼a zɛ-tsà ò-se¼lò

1.sg.sbj.pfv-take c6.sg-axe¼def IT -cut c2.sg-tree¼def
‘I took the axe to go cut the tree.’
The other non-nuclear SVCs also allow locational and temporal adverbials
between the verbs. However, in these cases there can be only one adverbial
and it has broad scope over the whole construction, as in example (40). This
suggests these SVCs share a single periphery and are joined via core co-
subordination. Nuclear SVCs do not allow locational or temporal adverbials
to occur between the verbs, consistent with their nuclear juncture.
(40) a. mà-dzɛ Òholò kivòe kí ̣ Akosua

1.sg.sbj-go Ho yesterday give Akosua
‘I went to Ho for Akosua yesterday.’
b. *mà-dzɛ Òholò kivòe kí ̣ Akosua òmonò
1.sg.sbj-go Ho yesterday give Akosua today
The distinguishing properties of these three subtypes of Avatime SVCs are

summarized in Table 24.2.
Table 24.2 Characteristic properties of Avatime SVC subtypes
Nuclear Core Sequential

Nuclear co- Core co- Core
Juncture–nexus type subordination subordination coordination
Can aspectual adverbials occur between verbs? Marginally Yes Yes

with restricted scope? No Yes Yes
Can the recurrent occur on subsequent verbs? No Yes Yes
with restricted scope? No No Yes
Can directionals occur on subsequent verbs? Marginally Yes Yes
with restricted scope? No No Yes
Can locational or temporal between verbs? No Yes Yes
adverbials occur with restricted scope? No No Yes
24.10 Semantic functions of SVCs and the Interclausal

Relations Hierarchy
24.10.1 The Interclausal Relations Hierarchy

There are striking tendencies across languages in how semantic functions
are expressed by syntactic constructions: closer semantic functions tend to
be expressed by syntactic constructions that more closely resemble a single
clause and have a stronger link between their elements, while looser

semantic relationships tend to be expressed by syntactic constructions with

a looser, or weaker, link between their elements. This is known as the
Interclausal Relations Hierarchy, which was first proposed by Silverstein
(1976), in his study of split case marking in ergative languages, as well as
by Givón (1980), in a more direct and detailed proposal. It has since been
further extended and solidified within the theory of RRG (Foley and Van
Valin 1984; Van Valin 2005). The hierarchy is presented in Figure 24.6.
The alignment of semantic functions and syntactic constructions is not
one to one. Not all languages make use of all the listed syntactic construc-
tion types. Each construction type may express multiple semantic functions
and, vice versa, a given semantic function may be expressible through
several different constructions. This leads to rather complex instantiations
of the interclausal relations hierarchies in individual languages (e.g. Casti
2012; Kockelman 2003) and a full consideration of the Avatime interclausal
relations is beyond the current paper. However, Avatime SVCs and related
semantic functions present some intriguing cases for this hierarchy, which
are worth discussing here. The primary prediction of the Interclausal Rela-
tions Hierarchy is that the tightest syntactic construction used to express a
semantic function should be tighter than or equal to the tightest syntactic
construction used to express a looser semantic function. There are two cases
involving SVCs where there is an apparent deviation from this prediction in
Avatime. By examining how Avatime SVCs fit within and deviate from the
Strongest Closest
Nuclear co-subordination Causative [1]
Phase
Nuclear subordination Manner
Daughter Motion
Peripheral Position
Nuclear coordination Means
Psych-action
Core co-subordination Purposive
Jussive
Core subordination Causative [2]
Daughter Direct perception
Peripheral Indirect perception
Core coordination Propositional attitude
Cognition
Clausal co-subordination Indirect discourse
Direct discourse
Clausal subordination Circumstances
Daughter Reason
Peripheral Conditional
Clausal coordination Concessive
Simultaneous actions
Sentential subordination Sequential actions
Situation-situation: unspecified
Sentential coordination
Weakest Loosest
Figure 24.6 The Interclausal Relations Hierarchy (Van Valin 2005: 209)

predictions of this hierarchy, we can gain a better understanding of the

hierarchy itself, Avatime SVCs, and their place within the wider ecology of
Avatime grammar.
24.10.2 Semantic Functions of Avatime SVCs

In this section, we review the semantic functions that can be expressed
using the three types of SVCs discussed in Section 24.9: nuclear co-
subordination, core co-subordination and core coordination. We will then
consider any alternative means of expressing these semantic functions in
Avatime and review the relations between syntactic constructions and
semantic functions in terms of the predictions of the Interclausal Relations
Hierarchy.
The nuclear co-subordination SVCs are used to express modifying sub-
events such as the position, (example (32)), or manner (example (33)) in
which an action is carried out. They are also used to express modification
of the path of motion, as in example (41). These three semantic functions
cluster neatly together in the Interclausal Relations Hierarchy and
together with means form the set of ‘modifying subevents’ functions
(Van Valin 2005: 206). Manner can also be expressed using a core SVC,
as discussed later.
(41) be-ple e-ku ní Gbàdzɛmɛ

c1.pl.sbj-descend svm-enter loc Gbadzeme
‘They descended into Gbadzeme.’ (Avatime-history_110905_BB_129)
As discussed in Section 24.5, it is also possible to express motion

accompanying an action with a directional prefix in a simple clause, as in
example (42).
(42) a-zɛ-bàsi¼bl c bà-li¼à

_ _
c1.sg.sbj.pfv-it-show¼1.pl.obj c5.pl-palm.tree¼def
‘He went to show us the palm trees.’ (Conv-ablorme_100715_SO-AS)
These directional prefix motion functions are a different type of motion

semantic function from the one expressed by the nuclear SVCs. The nuclear
SVCs further specify a path of motion. The directional prefixes express
concomitant motion which is either towards or away from the deictic
centre. This latter function is closer to the one described for the motion
semantic function in the hierarchy presented by Van Valin (2005: 206).
The use of directional prefixes to express motion creates an apparent
deviation from the predictions of the Interclausal Relations Hierarchy, since
clauses with directional prefixes are closer to simple clauses than the
nuclear SVCs used to express the tighter semantic function of manner. This
deviation likely arises from the derivation of the directional prefixes from
SVCs (see Section 24.4.2). If the directional prefixes were still verbs within
nuclear co-subordinate SVCs, these semantic functions would be expressed
by equally tight syntactic constructions and there would be no deviation

from the hierarchy. It is also notable that there is a semantic split between
the functions of the directional prefix and the motion nuclear SVCs where
this deviation occurs. It could then be possible to split motion into two
functions with accompanying motion positioned as closer than manner,
and path of motion remaining where motion is currently (looser than
manner). This would resolve the deviation in the hierarchy.
The core SVCs are used for adding and introducing or marking arguments.
Examples (34) and (40) show the use of this type of SVC to add a beneficiary
using the verb kí ‘give’. They can also be used to add instruments, manners
or means, as in examples (43)–(45), respectively, all with the verb k ̀ ‘take’. c
(43) a-k ̀
c kà-wɛ¼a yài ò-se¼lò
c1.sg.sbj-take c6.sg-axe¼def break c2.sg-tree¼def
‘He used an axe to split the tree.’
(44) a-k ̀
c ku-siyeyome sὲ
c1.sg.sbj.pfv-take c5.sg-anger leave
‘He left in anger.’
(45) a-k ̀
c ku-zò dzi ohonete
c1.sg.sbj.pfv-take c5.sg-theft become rich.person
‘Through theft he became a rich man.’
These SVCs can also be used to mark or introduce theme arguments which
are already called for by another verb in the SVC. For instance, in example
(46) k ̀ ‘take’ is used to introduce the theme banana. However, as example
c
(47) shows, kpɛ ‘put’ already licenses agent and theme arguments.
(46) a-k ̀
c k ̀ ranti¼ɛ
c kpɛ ní kà-s ¼ya
c mὲ
c1.sg.pfv-take banana¼def put loc c6.sg-basket¼def inside
‘S/he put the banana into the basket.’
(47) a-kpɛ lı̀-kùto¼lè

c1.sg.pfv-put c3.sg-hat¼def
‘S/he put the hat on.’
With the exceptions of manner and means, these functions are not
included within the list of interclausal semantic functions. These
argument-adding functions are relations between a predicate and its
dependents and are typically intraclausal. However, the fact that they
can be expressed by a combination of verbs in Avatime suggests they
could be added to the hierarchy. They bear clear similarities to the
causative [1] function – to directly bring about a state of affairs through
an event or action: for example, paint the table red or break the bowl. This
function also involves relations between a predicate and its arguments
and is sometimes expressed by a single verbal lexeme. Given the overlap
with manner and means functions and the general role of these func-
tions, it would make sense for them to be added immediately below
means in the hierarchy.

The core coordination SVCs are used to express sequential actions, as in

examples (35), (38) and (39). These SVCs can always be paraphrased by
coordinated clauses, as shown in example (48).
(48) a. ma-tsà tomatoes¼ye à-kpɛ ní

1.sg.sbj.pfv-cut tomatoes¼def svm.1.sg.pfv-put loc
kè-zi¼a mὲ
c6.sg-bowl¼def inside
‘I cut the tomatoes and put them in the bowl.’
b. ma-tsà tomatoes¼ye lɛ̌ mà-kpɛ ní
1.sg.sbj.pfv-cut tomatoes¼def and 1.sg.sbj.pfv-put loc
kè-zi¼a mὲ
c6.sg-bowl¼def inside
‘I cut the tomatoes and put them in the bowl.’
This use of core coordination to express sequential action is another

instance of a deviation from the predictions of the Interclausal Relations
Hierarchy. Sequential actions are among the loosest types of semantic rela-
tions. Many tighter semantic relations, such as circumstance and reason, are
expressed by the looser syntactic relations of clausal subordination and
coordination. For instance, the circumstantial in example (49), where a pre-
posed subordinate clause marked with the clause linkage marker xé indicates
the point in time at which the main clause happens.
(49) xé kù í-ṣ c p ́ ¼ɛ ki-tá-halı̀ ̣

c ̣
sı̀-wa¼sὲ
_ _
clm 1.pl.sbj.prog-sow finish¼cm 1.pl.sbj.pfv-int-gather c7-weed¼def
petee ní ̀ -ny ¼n ̀ mὲ
c c c
all loc c2.sg-farm¼def inside
‘When we finish sowing, we gather all the weeds from the farm.’
(rice_100613_EN-MM)
It is notable that, like with the previous deviation, a semantic distinction

not recognized in the hierarchy plays a role. Whereas all sequential actions
can be expressed by clausal coordination, the sequential actions that can be
expressed by core coordination SVCs must be performed by the same actor
and must constitute a single culturally relevant unit, generally with an
overarching goal (see Defina 2016b). If this type of sequential action relation
were distinguished as a separate, tighter, semantic relation category in the
hierarchy, the deviation would be resolved.
24.11 Summary
This part has surveyed the semantic functions and syntactic properties of
SVCs in Avatime, showing how they constitute three distinct juncture–
nexus types, see Table 24.3.
Each subtype of Avatime SVC poses a question for the Interclausal
Relations Hierarchy and its predicted links between syntactic

Table 24.3 Semantic functions of SVCs in each subtype
Subtype Semantic functions
Nuclear co-subordination Modifying (manner, motion (complex path), and position)

Core co-subordination Argument adding and theme marking
Modifying (manner and means)
Core coordination Sequential actions (same actor, coherent unit)
constructions and semantic functions. The argument-adding and marking

functions of the core co-subordination SVCs are not included within the
hierarchy, but should arguably be listed below the modifying functions.
The grammaticalization of SVCs indicating motion away from or towards
the deictic centre has created a deviation from the hierarchy’s predic-
tions, with the tighter directional prefix construction expressing a func-
tion that is looser than the manner function expressed by nuclear SVCs.
The use of core coordinating SVCs to combine certain types of sequential
actions presents another deviation. Both of these deviations relate only to
more restricted semantic functions and can be resolved through separat-
ing the specific semantic functions expressed with the tighter syntactic
construction from the more general motion and sequential action
semantic functions.
This review of Avatime SVCs within the context of the Interclausal Rela-
tions Hierarchy provides a clearer understanding of the internal structure of
Avatime SVCs and complex sentences more generally. It also suggests that
the list of semantic functions expressed by syntactic linkage is far from final
and may need further language-specific or general modifications or add-
itions. This is not an entirely new finding, and the point was made by Van
Valin (2005: 211) that the syntactic side of this hierarchy is better understood
than the semantic side, which should not be taken as an exhaustive list of
semantic functions.
References
Ameka, Felix K. 1995. The linguistic construction of space in Ewe. Cognitive

Casti, Francesco. 2012. Testing the Interclausal Relations Hierarchy: Aspectual and
Modal Periphrases in Modern Sardinian. PhD dissertation, University of
Manchester.
Caudal, Patrick and David Nicolas. 2005. Types of degrees and types of event
structures. In Claudia Maienborn and Angelika Wöllstein (eds.), Event
Arguments: Foundations and Applications, 277–300. Berlin: De Gruyter.
Defina, Rebecca. 2009. Aspect and Modality in Avatime. MA thesis, Leiden
University.

Defina, Rebecca. 2016a. Events in Languages and Thought: The Case of Serial Verb
Constructions in Avatime. PhD dissertation, Radboud University, Nijmegen.
Defina, Rebecca. 2016b. Serial verb constructions and their subtypes in
Avatime. Studies in Language 40(3): 648–680.
Defina, Rebecca. 2018. Tense, aspect and mood in Avatime. Afrika und Übersee
92: 65–97.
Dingemanse, Mark. 2012. Advances in the cross-linguistic study of ideo-
phones. Language and Linguistics Compass 6(10): 654–672.
Givón, T. 1980. The binding hierarchy and the typology of complements,
Kockelman, Paul. 2003. The Interclausal Relations Hierarchy in Q’ecqchi’
Maya. International Journal of American Linguistics 69: 25–48.
van Putten, Saskia. 2014a. Information Structure in Avatime. PhD dissertation,
Radboud University, Nijmegen.
van Putten, Saskia. 2014b. Left-dislocation and subordination in Avatime
(Kwa). In Rik van Gijn, Jeremy Hammond, Dejan Matić, Saskia van Putten
and Ana Vilacy Galucio (eds.), Information Structure and Reference Tracking in
Complex Sentences, 71–98. Amsterdam: John Benjamins.
van Putten, Saskia. 2016. Discourse functions of focus marking in Avatime.
Journal of African Languages and Linguistics 37(1): 91–130.
Schuh, Russell G. 1995. Avatime noun classes and concord. Studies in African
Linguistics 24(2): 123–149.
Dixon, ed., Grammatical Categories in Australian Languages, 112–171. Can-
Van Valin, Robert D., Jr. 1985. Case marking and the structure of the Lakhota
clause. In Johanna Nichols and Anthony C. Woodbury (eds.), Grammar
Inside and Outside the Clause, 363–413. Cambridge: Cambridge University
Press.
Van Valin, Robert D., Jr. 2005. Exploring the Syntax-Semantic Interface. Cam-
Welmers, William E. 1973. African Language Structures. Berkeley, CA: Univer-
sity of California Press.
Notes
1 All recordings are archived at The Language Archive at the MPI for
Psycholinguistics in Nijmegen (https://corpus1.mpi.nl/ds/asv/?1). Examples
in this chapter that are taken from this corpus are followed by the
filename of the recording.

2 This prefix is different from the habitual prefix, despite being homoph-
onous with it. See Defina (2018) for more details.
3 We could call this inalienable possession, but unlike most languages that
have an inalienable vs. alienable distinction, this class consists only of
kinship terms and does not include body parts.

25
of Amele (Papuan, Papua
New Guinea)
John R. Roberts
Abbreviations

A actor macrorole PrDP pre-detached position

ARG argument PRED predicate
ASP aspect PRO pronoun
CLM clause linkage marker PRSP prospective tense
CONT continuative aspect PSA privileged syntactic argument
CORE R RP core PSD possessed
DCA direct core argument PSR possessor
DEIC deictic QNT quantifier
DN DCA-non-macrorole R(EAL) realis status
DS different subject RG regret particle
DUn DCA-undergoer REMP remote past tense
DV dependent verb RP reference phrase
EVQ event quantification RPIP RP initial position
HORT hortatory RRG Role and Reference Grammar
HABP habitual past SEQ sequential event
IF illocutionary force SIM simultaneous event
operator
IU information unit SR switch-reference
INGR ingressive SS same subject
IRIT irregular iterative STA status operator
aspect
IT iterative aspect SVC serial verb construction
IVC impersonal verb TNS tense operator
construction
LS logical structure TODP today’s past tense

A Grammatical Sketch of Amele 915
MP modifier phrase U undergoer macrorole

NEGF negative future tense YESTP yesterday’s past tense
NMR non-macrorole - morpheme break
NUC clause nucleus <> infix
NUC R RP nucleus ¼ clitic break
NUM numeral ~ reduplication
PNG Papua New Guinea & sequential conjunction
PP postpositional phrase {&} seq conj in complex sentence
PoCS post-core slot ^ overlapping conjunction
PoDP post-detached position {^} overlap conj in complex
sentence
PrCS pre-core slot * ungrammatical
25.1 Introduction
Amele is a Papuan language spoken in Madang Province, Papua New Guinea

(PNG). The Amele people inhabit an area of approximately 120 square kilo-
metres between the Gum and Gogol rivers just south of the town of Madang in
PNG. The area extends from the coast to about 14 kilometres inland. Amele is
the largest of the Gum family of languages (Z’graggen 1975: 13) with a popula-
tion of approximately 5,300 speakers (Lewis et al. 2014).1
This chapter provides a linguistic description of the Amele language from
a Role and Reference Grammar (RRG) perspective. Section 25.2 describes the
basic syntax of the language and Section 25.3 examines serial verb construc-
tions and the switch-reference system.*
25.2 Common Topics
Nominative-accusative agreement is suffixed to the verb stem and up to four

core arguments can be marked on the verb. There are only two major lexical
categories, nouns and verbs, with very little overlap between them. Alterna-
tive undergoer selection may be made for ditransitive verbs. There is no
passive construction in the language and the only choice for privileged
syntactic argument (PSA) is [S, AT]. Focus may be expressed morphologically
and by incorporation of modifier elements into the verb word.
25.2.1 Basic Clause Patterns

Typologically, Amele has head-last syntax with OV order and postpositions.2
The language is also head-marking, and core arguments of the verb can be
* Robert Van Valin provided helpful comments and suggestions in the compilation of this chapter. However, any errors in
analysis or content are the responsibility of the author.

916 JOHN R. ROBERTS
PSA Adjunct Arg-Adjunct DCA VERB

(RP) (RP/PP) (RP/PP) (RP)
agent temporal goal patient
experiencer location path recipient
source perceived
instrument perceiver
benefactive addressee
accompaniment possessor
… …
Figure 25.1 Basic syntax of the clause in Amele
marked on the verb by cross-reference agreement. PSA agreement has one

basic type of morphology and direct core argument (DCA) agreement has
another basic type of morphology.3 The argument agreement morphology
follows a nominative-accusative pattern. Therefore, the PSA argument agree-
ment morphology is called nominative (nom) and the DCA agreement
morphology is called accusative (acc).
The basic syntax of the clause is given in Figure 25.1. The PSA is the first
element in the clause and the DCA occurs immediately preceding the verb.
After the PSA is the slot for any adjunct reference phrase (RP) or postpos-
itional phrase (PP). Following that is the slot where any argument-adjuncts
occur. Typically, no more than two or three non-verbal elements are
expressed in any one clause. However, up to four arguments can be encoded
on the verb.
There is also an order in which the different types of argument can be
marked on the verb. Up to three arguments can be marked on the verb and
the order is given in (1). The rightmost argument is the PSA agreement and
this is obligatory on the finite verb. This is the nominative agreement
morphology. The DCA-undergoer (DUn, where Un is undergoer) argument
agreement attaches directly to the verb stem. This is obligatory for some
verbs, optional for other verbs, and not allowed for others. Many verbs also
allow optional DCA-non-macrorole (DN) argument agreement to be marked
and this requires the applicative (applied object) marker. The form of the
DUn and DN marking is the accusative agreement morphology in each case.
It is possible to have a maximum of two DCA arguments marked on the verb,
either DUn þ DN or DN þ DN. The linear order of argument marking on the
verb in (1) is thus a mirror image of the ordering of arguments in the clause
in Figure 25.1.
(1) Order of arguments marked on the verb:

verb stem DUn.Agr APPLþDN.Agr (APPLþDN.Agr) þPSA.Agr
Verb agreement is formalized as in (2). nom agreement only applies to

the finite verb form. acc agreement can apply to both the finite and
infinitive verb forms, for example hel-ad-ec [throw-3pl.acc-inf] ‘to throw
them’, hel-i-ad-ec [throw-appl-3pl.acc-inf] ‘to throw to them’. See Roberts
(1996, 1997c).

(2) Verb agreement in Amele:

nom agreement cross-references the highest-ranking macrorole argument in
the finite verb.
acc agreement can cross-reference any other macrorole or non-macrorole
argument in the finite or non-finite verb.4
25.2.1.1 M-Intransitive Verbs

M-intransitive verbs have only one macrorole core argument. The main types
of M-intransitive verbs are stative verbs and motion verbs. There is no
copular verb like ‘be’ in Amele. Instead, the language uses several posture
verbs in certain stative predications (i.e. bilec ‘to sit (down)’, nijec ‘to lie
(down)’, tawec ‘to stand (up)’).5 As illustrated in Table 25.1, these posture
verbs have a state function and an activity function. For the state function
the single macrorole is undergoer. For the activity function the single
macrorole is actor. An example of each function is given in (3).
(3) Posture verb examples:

a. Jo jobon gemo taw-ena. State: posture
house village middle stand-3sg.nom.prs
be-village middle′ (stand′ (3sg [jo]))
‘The house stands in the middle of the village.’
b. Dana ben taw-im-ei ma-g-en. Activity: assume posture
man big stand-ss.seq-3sg.nom tell-1pl.acc-3sg.nom.remp
do′ (3sg [dana], [stand′ (3sg [dana])]) . . .
‘The headman stood up and spoke to us.’
c. Cam qila gagadic¼ca taw-ei-a. State: attributive
sun today strength¼add stand-3sg.nom-todp
today′ (be′ (3sg [cam], [strong′]))
‘The sun is strong today.’
There are two basic types of M-intransitive motion verb: those that specify
[moveþpath] and those that specify [moveþmanner]. [moveþpath]
motion verbs are verb-framed and [moveþmanner] motion verbs are
satellite-framed (Talmy 2007). Each type of motion verb is M-intransitive
and the single argument is actor in each case. However, there is a syntactic
difference between these types of motion verb. The [moveþpath] motion
Table 25.1 M-intransitive posture verbs
bilec ‘sit’ sit′ (x) state: posture

‘sit (down)’ do′ (x [sit′ (x)]) activity: assume posture
‘be’ be′ (x) state: attributive, identificational, specificational
nijec ‘lie’ lie′ (x) state: posture
‘lie (down)’ do′ (x [lie′ (x)]) activity: assume posture
tawec ‘stand’ stand′ (x) state: posture
‘stand (up)’ do′ (x [stand′ (x)]) activity: assume posture

918 JOHN R. ROBERTS
verb allows an allative argument to be expressed either with a ¼ca ‘towards’

PP, as in (4a), or with an applied object marked on the verb, as in (4b). The
applied object in (4b) is a non-macrorole direct core argument. The
[moveþmanner] motion verb, on the other hand, does not allow an allative
argument to be expressed, as illustrated in (5).
(4) Allative argument applicative alternation with [moveþpath] motion verbs:

a. Qa uqa¼ca b-ei-a.
dog 3sg¼towards come up-3sg.nom-todp
[do′ (3sg [qa], [move.upwards.towards.ref.point′ (3sg [qa])]) ^ proc
cover.path.distance′ (3sg [qa])] & ingr be-toward′ (3sg, 3sg [qa])
‘The dog came up to him.’
b. Qa b-i-t-oi-a.
dog come up-appl-3sg.acc-3sg.nom-todp
[do′ (3sg [qa], [move.upwards.towards.ref.point′ (3sg [qa])]) ^ proc
cover.path.distance′ (3sg [qa])] & ingr be-toward′ (3sg, 3sg [qa])
‘The dog came up to him.’
(5) Allative argument disallowed with [moveþmanner] motion verbs:

a. *Qa uqa¼ca cob-oi-a.
dog 3sg¼towards walk-3sg.nom-todp
b. *Qa cob-i-t-oi-a.
dog walk-appl-3sg.acc-3sg.nom-todp
Other verbs that are M-intransitive include ededec ‘gleam’ do′ (x, [gleam′ (x)]),
asalec ‘laugh’ do′ (x, [laugh′ (x)]), busuec ‘fart’ seml fart′ (x), silolec ‘ooze’ do′ (x,
[ooze′ (x)]), fojec ‘vomit’ do′ (x, [vomit′ (x)]), tatiec ‘look up’ do′ (x, [look.up′ (x)]),
bodoec ‘soften’ become soft′ (x), sanan mec ‘start’ ingr start′ (x).
Impersonal verbs also have to be treated as M-intransitive. A typical
example of an impersonal verb construction (IVC) is given in (6). There is
an optional free pronoun ija ‘1sg’ which corresponds to the subject pronoun
in the English translation. However, in the Amele form this pronoun is cross-
referenced on the verb with acc morphology. The nom agreement always
codes 3sg in an IVC but the reference is unspecified. The agreement is
dummy, neutral agreement which is part of the syntactic template for the
construction and there is no actor argument with these IVCs. Such IVCs
normally express a physiological or psychological experience and the 1sg.
acc agreement refers to the experiencer argument.
(6) Impersonal verb construction:

(Ija) cucui-t-ei-a.
(1sg) fear-1sg.acc-3sg.nom-todp
feel′ (1sg, [afraid′])
‘I am afraid.’
25.2.1.2 M-Transitive Verbs

M-transitive verbs have two macrorole core arguments: actor and undergoer.
Amele has M-transitive state verbs, such as fec ‘see’ see′ (x, (y)), doc ‘know’

know′ (x, (y)), gawec ‘want’ want′ (x, y), meleec ‘believe’ believe′ (x, (y)), cucuiec
‘fear’ fear′ (x, (y)). With some M-transitive non-state verbs, expression of the
undergoer argument on the verb is optional, with others this is obligatory.
(7) illustrates an active verb, j-ec [eat-inf] ‘to eat’, with optional DUn marking
on the verb. In (7a) this verb has the logical structure (LS) do′ (x, [eat′ (x, (y))])
with an optional second predicate argument. In (7b) this verb has the LS do′
(dana, [eat′ (dana, ha)]) with the second predicate argument position filled.
This argument is realized in the syntax by the RP ha ‘sugarcane’. Since ha is
neither animate nor a count noun, in this context it does not trigger acc
agreement on the verb. The predicate is, therefore, an activity with a non-
specific object. In (7c), on the other hand, ho ‘pigs’ is animate and a count
noun and therefore triggers acc agreement -ad ‘them’ on the verbs qoc ‘hit’
and jec ‘eat’.6 In this case the object is also specific and ‘eat’ is an active
accomplishment.
(7) Active verb with optional DUn marking:

a. Dana eu j-egi-na.
man that eat-3pl.nom-prs
do′ (3pl [dana], [eat′ (3pl [dana], (Ø))])
‘Those men are eating.’
b. Dana eu ha j-ein.
man that sugarcane eat-3pl.nom.remp
do′ (3pl [dana], [eat′ (3pl [dana], ha)])
‘Those men ate sugarcane.’
c. Dana eu age ho a-q-i je-ad-ein.
man that 3pl pig 3pl.acc-hit-dv eat-3pl.acc-3pl.nom.remp
[do′ (3pl [dana], [eat′ (3pl [dana], 3pl [ho])]) ^ proc consume′
(3pl [dana], 3pl [ho])] & [ingr consumed′ (3pl [ho])]
‘Those men killed the pigs and ate them.’
The M-transitive verb cesul-d-oc [help-3sg.acc-inf] ‘to help him/her’ in (8)

has obligatory DUn marking. In (8) the DUn marking -t ‘1sg.acc’ on the verb
expresses the obligatory second argument. The 3pl.nom argument is actor
and the 1sg.acc argument is undergoer.
(8) Active verb with obligatory DUn marking:

Age cesul-t-eig-a.
3pl help-1sg.acc-3pl.nom-todp
do′ (3pl, [help′ (3pl, 1sg])
‘They helped me.’
25.2.1.3 Three-Argument Verbs

Three-argument verbs have three core arguments in their LS. The ditransitive
verb ihac-d-oc [show-3sg.acc-inf] ‘to show him/her/it’ is illustrated in (9). The
LS of this verb is [do′ (x, Ø)] cause [become see′ (y, z)]. In (9a) the perceiver
argument 1sg is selected as undergoer and is realized on the verb as 1sg.
acc agreement attached directly to the verb stem. Because the perceiver
argument has been selected as undergoer, there is no way of encoding the

920 JOHN R. ROBERTS
perceived argument, ho eu ‘that pig’, on the verb. In (9b) ho eu ‘those pigs’ is

selected as undergoer and is realized on the verb as 3pl.acc agreement. In
this case, the 1sg.acc perceiver is also marked on the verb as a DN applied
object argument. (9c) shows that all three core arguments can be expressed as
agreement marking on the verb alone. Without any clarifying RPs in the
clause, (9d) is ambiguous between the first LS where 1sg is the perceiver
argument and the second LS where 1sg is the perceived argument –
although, the first interpretation would be deemed the more likely.
(9) 3-argument verb:
a. (Uqa) ho eu ihac-t-ei-a.
(3sg) pig that show-1sg.acc-3sg.nom-todp
[do′ (3sg, Ø)] cause [become see′ (1sg, ho)]
‘He showed me that pig.’
b. Ho eu ihac-ad-i-t-ei-a.
pig that show-3pl.acc-appl-1sg.acc-3sg.nom-todp
[do′ (3sg, Ø)] cause [become see′ (1sg, 3pl [ho])]
‘He showed those pigs to me.’
c. Ihac-ad-i-t-ei-a.
show-3pl.acc-appl-1sg.acc-3sg.nom-todp
[do′ (3sg, Ø)] cause [become see′ (1sg, 3pl)]
‘He showed them to me.’
d. Ihac-t-ei-a.
show-1sg.acc-3sg.nom-todp
[do′ (3sg, Ø)] cause [become see′ (1sg, Ø)] or
[do′ (3sg, Ø)] cause [become see′ (Ø, 1sg)]
‘He showed me something (unspecified).’ or ‘He showed me to someone
(unspecified).’
25.2.1.4 A Four-Argument Verb

There is one verb which allows an additional fourth core argument to be
marked. This is ‘give’, as illustrated in (10).7 Here the verb stem is realized as
acc morphology which agrees with the recipient. The meaning of ‘give’ is
constructional rather than lexical, and so there would be a constructional
schema for it. Furthermore, Latrouite and Van Valin (2014) argue that
constructional meaning should be represented differently from lexical
meaning; the constructional meaning for ‘give’ would be [do′ (x, Ø)] CAUSE
[INGR have′ (y, z)]. There is an infinitive form for each person and number
(e.g. itec ‘to give me’, ihec ‘to give you (sg)’, utec ‘to give him/her’, etc.) The
regular verb agreement morphology attaches to this derived stem. The first
argument (recipient) of have′ (x, y) in ‘give’ logical structure is assigned the
undergoer macrorole by virtue of being the argument encoded in the verb
stem itself. Thus, the recipient argument is always the undergoer and there
is no alternative construction where the second argument (theme) of have′
(x, y) is the undergoer. The ‘^ feel.negatively.affected′ (z)’ represents the
malefactive ‘on me’ argument. It is the LS proposed for the Japanese adversa-
tive passive (Imai 1998; Toratani 2002).

(10) The verb ‘give’:

Eeta¼nu ut-ad-i-t-ag-a?
what¼for 3sg.acc-3pl.acc-appl-1sg.acc-2sg.nom-todp
be-for′ (eeta, [do′ (2sg, Ø)] cause [INGR have′ (3sg, 3pl)] ^
feel.negatively.affected′ (1sg))
‘Why did you give him them on me?’
25.2.1.5 Non-Verbal Predicates

Amele does not have non-verbal predicates as such. With descriptive and
equational clauses a posture verb is used where it is necessary to specify a
clausal operator category. However, if such specifications are not required
then the posture verb is omitted. Examples are given in (11).
(11) Stative non-verbal predicates:

a. Mel eu hag¼ca. Attributive
boy that sickness¼with
‘That boy is sick.’
a′. Mel eu hag¼ca nij-en.
boy that sickness¼with lie-3sg.nom.remp
‘That boy was sick.’
be′ (mel, [sick′])
b. Misag uqa iwal-ad-ec. Identificational
Misag 3sg teach-3pl.acc-nmlz
‘Misag is a teacher.’
b′. Misag uqa iwal-ad-ec bil-ol-oi.
Misag 3sg teach-3pl.acc-nmlz sit-habp-3sg.nom
‘Misag used to be a teacher.’
be′ (Misag, [a teacher′])
c. Danben age wool-ad-ec. Specificational
Danben 3pl surpass-3pl.acc-nmlz
‘Danben (village) are the winners.’
c′. Danben age wool-ad-ec nij-ein.
Danben 3pl surpass-3pl.acc-nmlz lie-3pl.nom.remp
‘Danben (village) were the winners.’
be′ (Danben, [wooladec])
d. Mei ija¼na cof-t-ec. Equational
father.1sg.psr 1sg¼of supervise-1sg.acc-nmlz
‘My father is my boss.’
d′. Mei ija¼na cof-t-ec bil-en.
father.1sg.psr 1sg¼of supervise-1sg.acc-nmlz sit-3sg.nom.remp
‘My father was my boss.’
equate′ ([have.as.orientation.kin′ (1sg, mei-)], coftec)
25.2.2 Lexical Categories

Whereas English has four major word categories of verb, noun, adjective
and adverb, which can be distinguished on morphological and syntactic
grounds, Amele only has two: verbs and nouns. Words that function as

922 JOHN R. ROBERTS
nominal modifiers (adjectives in English) and verbal modifiers (adverbs in

English) cannot be distinguished on morphological or syntactic grounds in
Amele from words that function as the head of an RP (see Roberts 1987:
154–156, 158 for details). Therefore they are all categorized as ‘nouns’. In
English, many words can belong to multiple lexical categories, such as break
(verb and noun), red (noun and adjective), fast (adjective and adverb), round
(adjective, adverb, noun, verb, preposition). By contrast, in Amele there is
very little overlap between members of the verb class and those of the noun
class. Only three instances of overlap have been observed: cad ‘enemy’ (noun)
and cadec ‘to fight’ (verb), mele ‘truth’ (noun) and meleec ‘to believe’ (verb), and
cucuiec ‘to fear’ (verb) and cucuian ‘his/her fear’ (inalienably possessed noun).
25.2.2.1 Verbs
There are six morphosyntactic forms of the verb: regular verb, impersonal
verb, reciprocal verb, light verb, serial verb, and dependent switch-reference
verb. Serial verbs are described in Section 25.3.1. They are not marked for
tense or for nom agreement. Dependent switch-reference verbs are
described in Section 25.3.2. They are not marked for tense either. Amele
also has an interrogative verb adec ‘to when’, as illustrated in (12).
(12) Interrogative verb:

A~ad-eb uqa h-ugi-an?
dur ~when-3sg.nom.ds.sim.irr 3sg come-3sg.nom-fut
‘Whenever will he come?’
25.2.2.2 Nouns
Amele has three morphological classes of nouns: regular nouns with
uninflected stems, inalienably possessed nouns with possessor agreement
inflection, and deverbal nouns, derived from the infinitive form of a verb.
Inalienably possessed nouns are described in Roberts (1987: 171–175, 2015b).
They comprise kin terms,8 body-part terms and personal attribute terms.
Deverbal nouns are formed from the infinitive form of the verb. The infini-
tive suffix, -ec/-oc, on the verb functions as a nominalizing suffix on the noun
(e.g. cob-oc [walk-inf] ‘to walk’ or [walk-nmlz] ‘a walk’).
25.2.2.3 Postpositions
Postpositions are an important minor word class in Amele (Roberts 1987:
160–161). They are clitic words which must attach to a preceding host
element and they may not be stranded (Roberts 1991c, 1992, 1996). They
can be predicative or non-predicative, and we return to this in Section
25.2.4.2.
25.2.3 The Layered Structure of the Clause

In this section syntactic templates and operators are described and
illustrated.

SENTENCE
CLAUSE
(PERIPHERY)
CORE
(RP) (RP/PP) (RP/PP) (RP) (RP) NUC (ARG) (ARG) ARG
PRED
V DUn DN PSA
Figure 25.2 The layered structure of the clause
SENTENCE SENTENCE
PrDP CLAUSE CLAUSE PoDP
PrDP template PoDP template
CLAUSE CLAUSE
PrCS CORE CORE PoCS
PrCS template PoCS template

Figure 25.3 Optional syntactic structures
25.2.3.1 Syntactic Inventory

The basic structure of the clause in Amele is given in Figure 25.2 and Amele-
specific clause-internal linear precedence rules are given in (13). Amele has
pro-drop verb agreement which functions as the argument (ARG) of the
predicate. The only obligatory element in the active clause is the verb.
(13) Amele-specific clause-internal linear precedence rules:
a. XP* > CORE (verb final)
b. RP (PSA) > RP/PP (Adjunct) > RP/PP (Arg-Adjunct) > RP (DCA) > NUC
Templates for optional syntactic structures are given in Figure 25.3. The
pre-detached position (PrDP) is for topical established information and the
post-detached position (PoDP) is for additional information, such as clarifi-
cation. The pre-core slot (PrCS) is for focal new information or ‘heavy’
constructions, such as a nominal modified by a relative clause. The post-
core slot (PoCS) is for postposed clausal elements.
PrDP example
Example (14) illustrates a clause-external topic caja eu ‘that woman’ in the
pre-detached position. There is a resumptive pronoun uqa ‘she’ in the clause
for the argument in the PrDP.

924 JOHN R. ROBERTS
(14) Clause-external topic in PrDP:

Caja eu, uqa me qee.
woman that 3sg good not
‘That woman, she is no good.’
PoDP example
In (15) the RP dana eu ‘that man’ in the PoDP adds clarifying information to
the identity of the PSA eu ‘that’ in the clause. This pronoun is resumptive for
the argument in the PoDP.
(15) Clarifying information in PoDP:

Eu uqa jeje-g¼ca m-en, dana eu.
that 3sg voice-3sg.psr¼with put-3sg.nom.remp man eu
‘He got his voice back, that man.’
PrCS examples
In the unmarked form, temporal adjuncts (RP or PP) occur after the PSA RP,
as in (16a). Alternatively, the temporal adjunct can be placed in the PrCS as
focal new information, as in (16b).
(16) Temporal adjunct in PrCS:

a. Ija cum ceta gug filfil cabi¼na ceh-ig-an.
1sg yesterday yam kind different garden¼in plant-1sg.nom-yestp
‘I planted different types of yam in the garden yesterday.’
b. Cum ija ceta gug filfil cabi¼na ceh-ig-an.
yesterday 1sg yam kind different garden¼in plant-1sg.nom-yestp
‘Yesterday I planted different types of yam in the garden.’
Arguments and adjuncts with a relative clause are typically placed in the
PrCS, as shown in (17).
(17) Locative adjunct with a relative clause in the PrCS:

Cudun dan ben taw-ena eu¼na ija jahun-d-ug-a.
place fig big stand-3sg.nom.prs that¼at 1sg hide-3sg.acc-1sg.nom-todp
‘I hid it at the place where the big fig tree is.’
PoCS example
In (18) there is an instrumental argument-adjunct PP dubin¼na ‘with
stalk’ and a goal argument-adjunct PP camac ta¼na ‘into the sago scrapings’
in the final clause. The second argument-adjunct PP is located in the PoCS.
(18) Goal argument-adjunct PP in the PoCS:

Age wa wet-i dubin¼na basec-d-ogi-na camac ta¼na.
3pl water scoop-dv stalk¼with pour-3sg.acc-3pl.nom-prs sago scrapings¼in
‘They scoop up the water and pour it in with the stalk into the sago scrapings.’

25.2.3.2 Operators
Illocutionary force
Illocutionary force (IF) is the outermost operator and the form of expression
of different types of IF in Amele is illustrated in (19)–(23). Statement (19) is
expressed by a declarative sentence. Command (20) is expressed by the
imperative form of the verb. The imperative is identical in form to the
today’s past tense form. A yes/no question (21) is expressed with the
sentence-final question particle ¼fo. An information question (22) is
expressed by an interrogative word, such as in ‘who’, eeta ‘what’, cel ‘which’.
Exhortation (23) is expressed by the hortative form of the verb.
(19) IF: statement

Age aluh¼na bel-ein.9
3pl mountain¼to go.nsg-3pl.nom.remp
‘They went to the mountain.’
(20) IF: command

Age aluh¼na bel-eig-a!
2pl mountain¼to go.nsg-2pl.nom-imp
‘Go to the mountain!’
(21) IF: yes/no question

Age aluh¼na bel-ein¼fo?
3pl mountain¼to go.nsg-3pl.nom.remp¼q
‘Did they go to the mountain?’
(22) IF: information question

In aluh¼na nu-i-an?
who.sg mountain¼to go-3sg.nom-yestp
‘Who went to the mountain (yesterday)?’
(23) IF: exhortation

Ege aluh¼na bel-ec¼nu!
1pl mountain¼to go.nsg-inf¼hort
‘Let us go to the mountain!’
Evidentials
Amele does not have evidentials.
Status
Amele expresses realis/irrealis status on the switch-reference verb (see
Roberts 1990, 1994 for a wider study of this phenomenon in Papuan lan-
guages). The simultaneous ds nom agreement has different forms
depending on the realis/irrealis status of the clausal operator category
marked on the final verb in the linked clauses, traditionally called a clause
chain in Papuan linguistics. The clausal operator categories with realis
status are: present tense, today’s past tense, yesterday’s past tense, remote
past tense, habitual past tense/aspect, negative past tense. The clausal

926 JOHN R. ROBERTS
operator categories with irrealis status are: future tense, prospective (about
to) tense, negative future tense, and imperative, prohibitive, hortative,
optative, counterfactual, apprehensive IF categories. An example of realis
status is given in (24a) and an example of irrealis status is given in (24b).
(24) a. Marking of realis status:

Ho bu~busal-en age q-oig-a.
pig dur ~run out-3sg.nom.ds.sim.r 3pl hit-3pl.nom.todp
‘They killed the pig as it ran out.’
b. Marking of irrealis status:
Ho bu~busal-eb age q-oqag-an.
pig dur ~run out-3sg.nom.ds.sim.irr 3pl hit-3pl.nom-fut
‘They will kill the pig as it runs out.’
Tense
The tense categories in Amele are metrical. Metrical tenses mark degrees of
temporal remoteness from the deictic centre. Amele has a present tense (3a)
and three degrees of past tense: today’s past tense (3c), yesterday’s past tense
(22), and remote past tense (3b), as well as a negative past tense and a past
habitual tense/aspect (11b′).10 The language has a regular future tense (25), a
negative future tense (38), and a prospective (about to/intentional) (26)
tense.11 The event referred to in the prospective tense is nearer to the deictic
centre than the event referred to in the regular future tense. It is therefore
metrical.
(25) Future tense:

Age cabi¼na bel-oqag-an.
3pl going¼to go.nsg-3pl.nom-fut
‘They will go.’
(26) Prospective tense:

Age cabi¼na bel-oqag-a bil-i taw-eig-a.
3pl going¼to go.nsg-3pl.nom-prsp aux-dv stand-3pl.nom-todp
‘They stood about to go.’
Modality
Categories of deontic modality, such as ability, permission and obligation
are expressed lexically in Amele (Roberts 1987, 2001).
Event quantification (Roberts 2015a) is marked on the verb by distributive
inflection that is homonymous12 with the acc agreement, -ad ‘plural’ and -al
‘dual’. Some examples are given in (27). In (27a) the motion verb belec ‘to go’
is intransitive and the -ad ‘plural’ marker indicates a multiple event of each
woman going her own way. In (27b) and (c) calec ‘to arrive’ is intransitive. The
-ad ‘plural’ marker in (27b) indicates multiple events of men arriving, while
in (27c) the -al ‘dual’ marker indicates two events of men arriving.

(27) Plural and dual event quantification:

a. Caja age bud-u bel-ad-ein.
woman 3pl disperse-dv go.nsg-distr.pl-3pl.nom.remp
‘The women dispersed in all directions / each to her own place.’
b. Dana age cal-ad-ein.
man 3pl arrive-distr.pl -3pl.nom.remp
‘The men all arrived.’
c. Dana ale cal-al-esin.
man 3du arrive-distr.du-3du.nom.remp
‘The men both arrived.’
The EVQ marker -ad can also indicate an exclusive action, as in (28).
(28) Ija saen cecelac sum-i-h-ig-a qee¼nu

1sg time long wait-appl-2sg.acc-1sg.nom-todp not¼for
ija cu~cul-h-i l-im-ig nu-ad-ig-a¼da.
1sg dur~leave-2sg.acc-dv go-ss.seq-1sg.nom go-excl-1sg.nom-todp¼rg
‘I waited for you for a long time but in vain. So regretfully I left and went
without you.’
Aspect
There are three types of aspect that can be expressed by reduplicating some
part of the verb word: durative aspect, regular and irregular iterative aspect.
Durative aspect is marked on the simultaneous ss/ds verb and conveys
the idea that there is an extended temporal overlap of events. Without this
marking the temporal overlap is punctiliar. Durative aspect is expressed by
different kinds of CV~ or V~ reduplication of the verb stem (Roberts 1991a).
See (24a, b), (28), for examples of CV~ reduplication and (12) for an example
of V~ reduplication. With some verbs, the reduplicative (C)V~ marking of
durative aspect applies to the nom agreement suffixation rather than to the
verb stem, as illustrated by (29). If there is accusative marking on the verb
then the reduplicative marking for durative aspect applies there, as in (30).
(29) Durative aspect marked on nom agreement:

Co-Ø a-e~en
lips-3sg.psr open-dur ~ 3sg.nom.ds.sim.r
ija dunuh meci-d-ug-a.
1sg inside observe-3sg.acc-1sg.nom-todp
‘While he opened his mouth I looked inside.’
(30) Durative aspect marked on acc agreement:

Age eu cunug ihac-te~t-eig
3pl that all show-dur ~ 1sg.acc-3pl.nom.ss.sim
sa-t-ein.
explain-1sg.acc-3pl.nom-remp
‘As they showed me everything they explained it to me (lit. explained me).’
The meaning of the regular iterative is a repeated, regular action. This

aspect is expressed by rightward reduplication of the whole stem if the verb

928 JOHN R. ROBERTS
does not have an acc marker, otherwise the acc marker is reduplicated
either in place of, or in addition to the reduplication of the verb stem
(Roberts 1987: 252–256). For some minimal stem verbs, such as l-ec ‘to go’,
the reduplicated stem is of the serial verb form (e.g. li~li-ec ‘to go repeatedly’;
see Section 25.3.2). An example of regular iterative aspect is given in (31a).
(31) a. Regular iterative:

Gow-ec eu fale~fale-ei-a.
light-nmlz that flash~it-3sg.nom-todp
〈IF DECL 〈TNS TODP 〈ASP IT seml flash′ (gowec) 〉〉〉
‘That light flashed repeatedly.’
b. Irregular iterative:
Gow-ec eu fale~fule-ei-a.
light-nmlz that flash~irit-3sg.nom-todp
〈IF DECL 〈TNS TODP 〈ASP IRIT seml flash′ (gowec) 〉〉〉
‘That light flashed intermittently.’
The meaning of the irregular iterative is a repeated action that is irregular

in some way (i.e. haphazard, spasmodic, intermittent, etc.). This form
involves reduplication of the verb stem but with a disharmonic vowel
change in the reduplicated formant. There are eight types of disharmonic
vowel change possibilities, which are determined by phonological factors.13
An example of irregular iterative is given in (31b).
The aspectual notions of continuative and completive can be expressed
periphrastically with a serial verb construction (see examples (45) and (46) in
section 25.3.1).
25.2.4 The Structure of RPs and PPs

25.2.4.1 RP Structure
The various structures of the RP are given in (32). The head noun in the RP
can be a regular noun, an inalienably possessed noun or a deverbal noun.
The modifier can be a noun, such as dana caub [man white] ‘white man’, dana
me [man good] ‘good man’, or a modifier (a small class of words that only
function as modifiers), such as dana bahic [man very] ‘real man’, dana qee
[man not] ‘not a man’, or a phrase, such as dana mel iwal-ad-ec [man child
teach-3pl.acc-nmlz] ‘school teacher’.
(32) Reference phrase structures:

a. noun modifier (numeric) quantifier deictic/indefinite article
universal quantifier
b. interrogative deictic þ noun
c. PP/RP þ noun
A mass noun can be modified by a general quantifier, such as leih ‘some’ or

geh ‘much’. A count noun can be modified by a numeric quantifier, such as
lecis ‘two’ or cijed ‘three’. Amele has a pental counting system and the
cardinal numbers are: osahic/osol ‘one’, lecis ‘two’, cijed ‘three’ wal oso ‘four’,

ebum oso ‘one hand (10)’. Ordinal numbers can be formed by adding the
nominalizing suffix -doc to the numeral (e.g. osahic-doc ‘first’, lecis-doc ‘second’,
cijed-doc ‘third’). There are also a number of relationship terms that can
function as ordinal numbers, for instance matu ‘firstborn/first’, milum
‘secondborn/second’, subig ‘lastborn/last’.
The quantifier can be followed by a deictic element, such as eu ‘that’ or
ceheleg ‘up there’, or the indefinite article oso ‘a/one’. The article can co-occur
with a deictic element, as in dana eu oso [man that art] ‘one of those men’.
The universal quantifier cunug ‘all’ occurs at the end of the RP, as in jo nag
cijed eu cunug [house small three that all] ‘all those three small houses’.
With respect to (32a), the head noun can be questioned with an interroga-
tive, such as eeta ‘what’ or in ‘who’, for example Hina ija-in in? [2sg name-2sg.
psr who.sg] ‘What (lit. who) is your name?’ The numeric quantifier can be
questioned with an interrogative, such as ganic ‘how much/many’, for
example jo ganic? [house how many] ‘how many houses?’, saab ganic? [food
how much] ‘how much food?’. However, when the deictic is questioned, the
interrogative form, such as cel ‘which’, is placed in the RPIP, which is the pre-
core slot in the RP, as in (32b). Thus the interrogative form is cel dana [which
man] ‘which man?’.
With respect to (32c), the preceding PP/RP expresses the possessor of the
noun or a specification of the noun. (33) illustrates the difference in how
alienable and inalienable possession is expressed.14 Alienable possession in
(33a) is expressed with a possessive PP preceding the possessed noun. The
possessive PP functions as an argument of the possessed noun, which is the
head of the nucleus of the possessive noun phrase. The possessive PP is
therefore non-predicative. Inalienable possession in (33b) is expressed by
suffixal agreement in person (first, second or third) and number (singular,
dual or plural) with the possessor noun phrase. The possessor need not be
expressed by an overt RP or pronoun. Thus the possessor agreement func-
tions as the argument (ARG) of the inalienably possessed head noun.15
Semantically, inalienably possessed nouns are kinship terms, body-part
terms or personal attributes. In (33b) cuduni means ‘my personal place’ or
‘the place that belongs to me in some way’, whereas in (33a) ijana cudun
simply means ‘my place’ without the connotations of personal ownership.
(33) a. Alienable possession:

ija¼na cudun ‘my place’
1sg¼of place
have′ (1sg, cudun)
b. Inalienable possession:
(ija) cudu-ni ‘my (personal) place’
1sg place-1sg.psr
have.as.attribute′ (1sg, cudu-)
When the modifier follows the head noun it has an attributive function.
In (34a) jo us nijec describes a house where people are sleeping. When the

930 JOHN R. ROBERTS
modifier precedes the head noun it has a specificational function. In (34b) us

nijec describes the type of house. Adjectival-cum-nominal modifiers are
represented as predicates which take the item in the NUCR as an argument
(underlined).
(34) a. Modifier with attributive function:

jo us nij-ec ‘sleeping house (i.e. a house that is asleep)’
house sleep lie-nmlz
be′ (jo, [asleep′])
b. Modifier with specificational function:
us nij-ec jo ‘sleeping house (i.e. a house for sleeping in)’
sleep lie-nmlz house
be′ (jo, [PURP sleep′])
The basic structure of the RP is given in Figure 25.4. Most modifying

elements in the RP can occur on their own as referring expressions. Some
examples are given in (35). Thus, such modifiers are represented in the
syntactic structure as well as in the operator projection.
(35) a. Ben/eu/oso h-ona.

big/that/one come-3sg.nom.prs
‘The big (man) is coming.’/ ‘That (man) is coming.’/ ‘Someone is coming.’
b. Cijed h-ogi-na.
three come-3pl.nom-prs
‘The three (men) are coming.’
The structure of the alienably possessed PP and inalienably possessed RP

are given in Figure 25.5. Ija cebinami ‘my brother’ is the inalienably possessed
RP with cebinami as the head noun. Caja ‘woman’ is the head noun of the
alienably possessed RP and ija cebinamina functions as the possessor PP. It is
clear from the reference that ija ‘1sg’ is a woman.
RP MP
CORER MP
NUCR MP
N N NUM DEM
jo nag cijed eu ‘those three small houses’

house small three that
NUCR
CORER QNT
RP DEF
RP DEIC
Figure 25.4 The basic structure of the RP in Amele

RP
CORER
PP NUCR
RP P
RPIP CORER
RP NUCR ARG
PRO N N
ija cebina-mi =na caja ‘my brother’s woman’

1SG opp.sex.sibling-1SG.PSR =of woman
′ (have.as.procreation. ′ (1SG, cebina-), caja)
Figure 25.5 The structure of the possessive RP in Amele
Amele uses a pronominal copy strategy where an RP is immediately

followed by a personal pronoun. The pronoun indicates the person and
number of the RP, as illustrated in (36). The pronoun functions as appositive
to the RP and since it is not possible for other clausal elements to occur
between the RP and the pronoun, it forms a constituent with the RP.
(36) Pronominal copy strategy:

Caja eu uqa/ale/age qaj-ei/-esi/-eig-a.
woman that 3sg/3du/3pl cry-3sg.nom/3du.nom/3pl.nom-todp
‘That/those (du)/those (pl) woman/women cried.’
25.2.4.2 PP Structure
Amele only has postpositional phrases (PPs). They can be predicative or
non-predicative. Predicative PPs occur with temporal or locative adjuncts
(e.g. Mande¼na [Monday¼on] ‘on Monday’, be-on′ (Mande, x) or jo¼na
[house¼in] ‘in the house’ be-in′ (jo, x)), or with goal, path, source, or
instrument argument-adjuncts, as described in Section 25.2.7. The
possessor-marking postposition ¼na ‘of’, described in Section 25.2.4.1, is
non-predicative.
25.2.5 Constraints on A and U Selection

With a ditransitive verb like ihacdoc ‘to show him/her’ in (9), either the
perceiver x-argument in see′ (x, y) or the perceived argument y can be
selected as undergoer. The latter is the unmarked choice. When the per-
ceived argument is selected as undergoer it is coded on the verb with acc
marking attached directly to the verb stem and the perceiver argument can
also be marked as a non-macrorole applied object. However, if the perceiver
is selected as undergoer it is coded on the verb with acc marking attached

932 JOHN R. ROBERTS
directly to the verb stem and the perceived argument cannot be marked on
the verb.
The verb ‘give’ is exceptional in that the undergoer argument func-
tions as the stem of the verb. The first argument of have′ (x, y) (recipient)
in ‘give’ LS is thus the only choice for undergoer assignment and there is
no alternative argument-marking construction where the second argu-
ment of have′ (x, y) (theme) is the undergoer. An example of this was
given in (10).
When the goal argument in an argument-adjunct PP is a person, an
applied object construction may be used as an alternative expression, as
illustrated in (4). This is also the case for the source adjunct PP argument,
and the benefactive adjunct PP argument, illustrated in (37). For the mal-
efactive argument shown in (38) there is only the applicative form and there
is no corresponding argument-adjunct PP form.
(37) Benefactive argument-adjunct applied object:

a. Uqa age¼nu jo eu ceh-al-ei-a.
3sg 2pl¼for house that plant-3du.acc-3sg.nom-todp
‘He built both those (two) houses for you (plural).’
b. Uqa jo eu ceh-al-i-ad-ei-a.
3sg house that plant-3du.acc-appl-2pl.acc-3sg.nom-todp
‘He built both those (two) houses for you (plural).’
c. [do′ (3sg, [build′ (3sg, 3du [jo])]) ^ proc create′ (3du [jo])] & ingr exist′
(3du [jo]) purp [become have′ (2pl, 3du [jo])]
(38) Malefactive argument-adjunct:

Ene cain salal-i-t-ag-aun.
here proh slide-appl-1sg.acc-2sg.nom-negf
NOT [be-here′ (2sg, [do′ (2sg [slide′ (2sg)]] ^ feel.negatively.affected′ (1sg)]
‘Don’t slide here and annoy me (coll. don’t slide on me here).’
25.2.6 Postposition Assignment

The only non-predicative postposition assigned by rule is ¼na, which is
assigned to the possessor in the alienable possession RP (cf. 33a).
(39) Rule assigning ¼na ‘possessor’:

Assign ¼na to the x-argument in the RP logical structure segment:
have′ (x, y)
25.2.7 Coding of Adjuncts

Temporal adjuncts can be an RP, such as cum ‘yesterday’, uqadec ‘tomorrow’,
qila ‘today/now’, cel saen ‘what time’, or a PP, such as Mande¼na ‘on Monday’,
Ogas¼na ‘in August’. Locative adjuncts can be an RP, such as ene ‘here’, ono
‘there’, ceheleg ‘up there’, ana ‘where’, or a PP, such as jo¼na [house¼in] ‘in
the house’, na¼na [tree¼on] ‘on the tree’, eeta¼na [what¼in] ‘in what’.

25.2.8 PSA Alignment(s)

In Amele the only choice for PSA with a transitive verb is [AT] as there is no
passive construction in the language. However, IVCs have exceptional PSA
verb agreement coding. Although the undergoer is the only DCA in an IVC
and therefore the highest-ranking DCA, it is coded as acc, contrary to (2).
This is because 3sg.nom is assigned by default in an IVC.
25.2.9 Information Structure

The pre-detached position is outside the potential focus domain. The
information structure of (14) is illustrated in Figure 25.6. The RP caja eu
‘that woman’ is a clause-external topic in the pre-detached position and,
as a topic, it is outside of the potential focus domain of the proposition.
Uqa ‘she’ is the resumptive pronoun in the clause. The potential focus
domain (dotted line) is the attributive clause uqa me qee ‘she (is) no good’
and the actual focus domain (solid line) is the predication me qee ‘(is)
no good’.
Similarly, the post-detached position is outside the potential focus
domain. The information structure of (15) is illustrated in Figure 25.7.
The RP dana eu ‘that man’ is in the post-detached position and clarifies
the identity of eu ‘that’ in the main clause. As clarifying information, it is
outside of the potential focus domain of the proposition. The potential
focus domain is the statement eu uqa jejegca men ‘he got his voice back’.
Here eu ‘that (man)’ functions as a resumptive pronoun for dana eu ‘that
SENTENCE
PrDP CLAUSE
CORE
RP PRO NUC NEG
PRED
‘That woman, she is no good.’
N
(14) Caja eu, uqa me qee.

woman that 3SG good not
IU IU IU
SPEECH ACT
Figure 25.6 The PrDP and the potential focus domain

934 JOHN R. ROBERTS
SENTENCE
CLAUSE PoDP
RP CORE
PP NUC ARG RP
PRED
PRO V
(15) Eu uqa jeje-g=ca m-en, dana eu.

that 3SG voice-3SG.PSR=with put-3SG.NOM.REMP man eu
IU IU IU IU
‘He got his voice back, that man.’
SPEECH ACT
Figure 25.7 The PoDP and the potential focus domain
man’. The actual focus domain is the predication uqa jejegca men ‘got his
voice back’.
Amele has a range of particles that can attach to the end of the sentence
which qualify the proposition, such as da ‘but, however, nevertheless’, do ‘be
encouraged, let’s do it’, fo ‘yes/no question’ fa ‘dubitive question, maybe’, ijom
‘certainly’, le ‘permission granted’, and mo ‘supplication, pleading’ (Roberts
1990). Some of the sentence particles, such as the question particles, fo and fa,
can occur either at the end of the sentence or they can be focused on a
particular constituent in the sentence. In (40a) the yes/no question particle,
fo, occurs at the end of the sentence. It has scope over the whole proposition
and therefore expresses sentence focus. In (40b–c) the question particle is
focused on a particular sentence constituent and expresses marked narrow
focus on that constituent. (40d) illustrates the use of the question word in
‘who’. It is ungrammatical to have the question particle fo at the end of the
sentence in this case since the question word expresses narrow focus and the
particle at the end of the sentence would express sentence focus. However, as
illustrated in (40e), it is possible to focus on the question word with the
question particle. This expresses marked narrow focus.
(40) Focus expressed morphologically:

a. Ija aide-ni cabi¼na nu-i-a¼fo? [unmarked sentence focus]
1sg wife-1sg.psr garden¼to go-3sg.nom-todp¼q
‘Did my wife go to the garden?’
b. Ija aide-ni¼fo cabi¼na nu-i-a? [marked narrow focus]
1sg wife-1sg.psr¼q garden¼to go-3sg.nom-todp
‘Was it my wife that went to the garden?’

c. Ija aide-ni cabi¼na¼fo nu-i-a? [marked narrow focus]

1sg wife-1sg.psr garden¼to¼q go-3sg.nom-todp
‘Was it to the garden that my wife went?’
d. In cabi¼na nu-i-a(*¼fo)? [unmarked narrow focus]
who.sg garden¼to go-3sg.nom-todp(¼q)
‘Who went to the garden?’
e. In¼fo cabi¼na nu-i-a? [marked narrow focus]
who.sg¼q garden¼to go-3sg.nom-todp
‘Who is it that went to the garden?’
It is possible to incorporate certain modifier constituents into the verb

word to express emphatic focus for the purposes of contrast or correction, or
closer specification, for instance. For example, the intensifier bahic ‘very,
must, really’ can occur either preceding the verb, as in (41a), or be incorpor-
ated into the verb between the verb stem and the verb suffixation, as in
(41b). Other modifiers which can be focused in this way are the limiters dih
‘just’ and himec ‘only’, and the negators qee ‘not’ and cain ‘don’t’.
(41) Emphatic focus:

a. Age Anut bina-n bahic sul-eig-a!
2pl God fame-3sg.psr really lift up-2pl.nom-imp
‘Really praise God!’
b. Age Anut bina-n sul<bahic>eig-a!
2pl God fame-3sg.psr lift up<really>2pl.nom-imp
‘REALLY praise God!’
25.2.10 Linking
In this section the semantics-to-syntax linking for some two-place and three-
place predicates is illustrated. Figure 25.8 shows the semantics-to-syntax
linking in a transitive sentence with acc (DUn) marking. Here 3pl [dana]
is the first argument in the active predicate LS do′ (3pl, [eat′ (3pl [dana], 3pl
[ho])]), and is assigned the actor macrorole. 3pl [ho] ‘pigs’ is the second
argument in this LS and also the only argument in the stative predicate LS
ingr consumed′ (3pl [ho]). This argument is assigned the undergoer macro-
role. The actor argument, 3pl [dana], is selected as PSA and coded as nom
agreement on the verb. The undergoer argument, 3pl [ho], is coded as
acc agreement on the verb in order to indicate the plurality of the
animate referent.
Figure 25.9 illustrates the semantics-to-syntax linking in an IVC. The 1sg
argument in feel′ (1sg, [afraid′]) is assigned the undergoer macrorole
according to the Actor-Undergoer Hierarchy (AUH). The IVC is coded as
3sg.nom agreement by default. There is therefore no linking between this
argument marking in the syntax and the logical structure.
Figure 25.10 illustrates the semantics-to-syntax linking for a three-place
predicate, ihacdoc ‘to show him’. The leftmost argument, 3sg ‘he’, is
assigned the actor macrorole and the rightmost argument, 3pl [ho] ‘pigs’,

936 JOHN R. ROBERTS
SENTENCE
CLAUSE
RP RP CORE
RP RP NUC ARG ARG
PRED
‘Those men … ate the pigs.
N PRO N V
(7c) Dana eu age ho … je-ad-ein.

man that 3PL pig eat-3PL.ACC-3PL.NOM.REMP
PSA: NOM ACTIVE: 3PL DCA: ACC
ACTOR UNDERGOER
[ ′ (3PL [dana], [ ′ (3PL [dana], 3PL [ho])]) … ] & [INGR ′ (3PL [ho])]
Figure 25.8 Semantics-to-syntax linking in a transitive sentence with ACC marking
SENTENCE
CLAUSE
RP CORE
NUC ARG
PRED
‘I am afraid.’
PRO V
(6) Ija cucui-t-ei-a.

1SG fear-1SG.ACC-3SG.NOM-TODP
IMPERSONAL: 3SG.NOM PSA: ACC
UNDERGOER
feel′ (1SG, [afraid′ ])

Figure 25.9 Semantics-to-syntax linking in an impersonal verb construction
is assigned the undergoer macrorole. The actor is selected as PSA and coded
as nom agreement on the verb. The undergoer is coded as a DUn by acc
agreement attaching directly to the verb stem. The non-macrorole core
argument, 1sg ‘me’, is coded as a DN by the applicative marker and acc
agreement marking on the verb.

SENTENCE
CLAUSE
RP CORE
NUC ARG ARG ARG
PRED
‘He showed those pigs to me.’
N V
(9b) Ho eu ihac-ad-i-t-ei-a.
pig that show-3PL.ACC-APPL-1SG.ACC-3SG.NOM-TODP
PSA: NOM ACTIVE: 3SG DCA: ACC DCA: ACC
ACTOR NMR UNDERGOER
[ ′ (3SG, Ø)] CAUSE [BECOME ′ (1SG, 3PL [ho])]

Figure 25.10 Semantics-to-syntax linking in a ditransitive predicate
25.3 Language-Specific Topics
The language-specific topics discussed in this chapter are serial verb con-
structions in Section 25.3.1 and switch-reference in Section 25.3.2.
25.3.1 Serial Verb Constructions

Kroeger (2004: 229–230) shows that prototypical serial verb constructions
(SVCs) have the following syntactic and semantic properties:16
(42) Characteristic properties of SVCs:

a. A prototypical SVC contains two or more morphologically independent
verbs within the same clause, neither of which is an auxiliary.
b. There are no conjunctions or other overt markers of subordination or
coordination separating the two verbs.
c. The serial verbs belong to a single intonation contour, with no pause
separating them.
d. The entire SVC refers to a single (possibly complex) event.
e. A true SVC may contain only one specification for tense, aspect, modality,
negation, etc., though these features are sometimes redundantly marked
on both verbs.17
f. The two verbs in the SVC share at least one semantic argument.

938 JOHN R. ROBERTS
g. Obligatory non-coreference: a true SVC will not contain two overt NPs
which refer to the same argument.
h. A prototypical SVC contains only one grammatical subject.
SVCs in Amele do not have all of the prototypical properties listed in (42).
In an SVC the non-final verb in the series is marked with -i or -u. Verbs with
an -ec infinitive form take -i, while verbs with an -oc infinitive form take -u.
The -i/-u marking indicates the verb is dependent (dv).18 This contravenes
principle (42b) as the dv marker indicates a nexus relationship. The PSAs of
the verbs in an SVC typically have the same referent.19 There are two basic
types of SVC. In one type, illustrated in (43a), the verbs are in a cosubordi-
nate relationship. In the other type, illustrated in (44a), the verbs are in a
superordinate–subordinate relationship. By comparison, the verb manimei in
(43b) is fully inflected with switch-reference morphology and is therefore
not a serial verb. Similarly, ehimeig in (44b) is not a serial verb.
(43) Cosubordinate SVC:

a. Caja uqa ceta man-i j-ei-a.
woman 3sg yam roast-dv eat-3sg.nom-todp
〈IF DECL 〈STA REAL 〈TNS TODP [[do′ (3sg [caja], Ø)] cause [become roasted′
(ceta)] {&} [[do′ (3sg [caja], [eat′ (3sg [caja], ceta) ^ proc consume′
(3sg [caja], ceta)] & [ingr consumed′ (ceta)]])] 〉〉〉
‘The woman roasted and ate yam.’20
b. Caja uqa ceta man-im-ei j-ei-a.
woman 3sg yam roast-ss.seq-3sg.nom eat-3sg.nom-todp
〈IF DECL 〈STA REAL 〈TNS TODP [do′ (3sg [caja], Ø)] cause [become roasted′
(ceta)] {&} [[do′ (3sg [caja], [eat′ (3sg [caja], ceta) ^ proc consume′
(3sg [caja], ceta)] & [ingr consumed′ (ceta)]])] 〉〉〉
‘The woman roasted yam and ate it.’
(44) Subordinate SVC:

a. Age ja eh-i n-eig-a.
3pl firewood take-dv come down-3pl.nom-todp
〈IF DECL 〈STA REAL 〈TNS TODP 〈DIR COME DOWN [[do′ (3pl, Ø)] cause [do′ (3pl,
[move.away.from.ref.point′ (ja)])] 〉〉〉〉
‘They brought firewood down.’
b. Age ja eh-im-eig n-eig-a.
3pl firewood take-ss.seq-3pl.nom come down-3pl.nom-todp
〈IF DECL 〈STA REAL 〈TNS TODP [do′ (3pl, Ø)] cause [do′ (3pl, [move.away.
from.ref.point′ (ja)])] {&} [do′ (3pl, [move.down.to.ref.point′ (3pl)])] 〉〉〉
‘They brought firewood and came down.’
In (43a) the verbs mani ‘roast’ and jeia ‘she ate’ describe a series of closely
related events. Compare (43b) in which manimei is the fully inflected ss.seq
form. Here the roasting and eating are interpreted as separate consecutive
events and they are separate clauses in the syntactic representation. The
arguments caja ‘woman’ and ceta ‘yam’ in (43a) are shared in the LS for
the two predicates. In the syntax for (43a) illustrated in Figure 25.11 the
argument caja is represented by an RP in the clause and by 3sg.nom

SENTENCE
CLAUSE
RP CORE
CORE CORE
RP RP RP NUC NUC ARG
PRED PRED
N PRO N V V ‘The woman roasted

and ate yam.’
(43a) Caja uqa ceta man-i j-ei-a.
woman 3SG yam roast-DV eat-3SG.NOM-TODP
Figure 25.11 SVC with cosubordinate core juncture
agreement on the verb jeia ‘she ate’. The argument ceta is represented as a
core RP argument of mani ‘roast’. It is not coded on the verb by acc
agreement as it is inanimate and a mass noun. The SVC is a cosubordinate
CORE [[CORE] [CORE]] juncture because it is not possible to have temporal or
locative clausal adjuncts applying separately to either core. By comparison,
in (43b) it would be possible to qualify jeia ‘she ate’ with the temporal
adjunct hibna ‘later’ because it is a separate clause.
(44a) describes a single event of ‘take down’. The verb n- ‘come down’ is a
directional modifier of the verb ehi ‘take’ and is coded in the LS as such.
There is no shared argument between the verbs as n- is represented in the LS
as a directional operator. (44a) therefore contravenes principle (42f). Com-
pare (44b), where ehimeig is fully inflected for ss.seq. Here the event ehimeig
‘they took’ is linked to the event neiga ‘they came down’ by {&} in the LS.
They are separate consecutive events in the LS and are separate clauses in
the syntactic structure. In the syntactic structure for (44a) in Figure 25.12 the
verb n- is in a subordinate relationship to ehi. This is ad-nuclear
subordination. The 3pl.nom argument agreement -eig attaches to n- as the
final verb in the series.
With respect to the characteristic SVC properties detailed in (42), a par-
ticular SVC may not necessarily belong to a single intonation contour
(cf. 42c). Compare the sentence in Figure 25.13. The li ‘go’ verb is a direc-
tional modifier of ehi ‘take’. The first two cores form one intonational unit
and the last core is a separate intonational unit. However, the dv coding
marks this as describing a series of linked events which should be inter-
preted as a unified complex event.
The subordinate SVC has a range of modifying functions in addition to
expressing directionality. Examples are given in (45)–(49). In each case the
modifying verb follows the verb that is modified. (45)–(46) express aspect
periphrastically. (45) illustrates how continuative aspect is expressed by a

940 JOHN R. ROBERTS
SENTENCE
CLAUSE
RP CORE
RP NUC NUC ARG
PRED
‘They brought the firewood down.’
PRO N V V
(44a) Age ja eh-i n-eig-a.

3PL firewood take-DV come down-3PL.NOM-TODP
Figure 25.12 SVC with nuclear subordination
SENTENCE
CLAUSE
RP CORE CORE CORE
RP RP RP NUC NUC NUC RP NUC ARG
PRED PRED PRED
N PRO N V V V N V
Caja age saab tac-i eh-i l-i bisin ohis m-egi-na.

woman 3PL food fill up-DV take-DV go-DV store above put-3PL.NOM.PRS
‘The women fill up (their bilums) with food, take it away and put it in the attic store.’
Figure 25.13 Extended coordinate SVC
posture verb: bilec ‘to sit’, nijec ‘to lie’, tawec ‘to stand’, and (46) illustrates
how completive aspect is expressed with the verb hedoc ‘to finish it’. (47)–(49)
illustrate some other common forms of this type of subordinate SVC.
(45) Continuative aspect expressed with a posture verb:

Age nu-i bil-egi-na.
3pl go-dv sit-3pl.nom-prs
‘They go continuously.’
(46) Completive aspect expressed with the verb hedoc ‘to finish it’:
Age jo ceh-i he-d-oig-a.
3pl house plant-dv finish-3sg.acc-3pl.nom-todp
‘They finished building the house.
(47) Investigative modification expressed with the verb fec ‘to see’:
Uqa wehuc j-i f-ei-a.
3sg soup eat-dv see-3sg.nom-todp
‘She tasted the soup.’

(48) Superlative modification expressed with the verb cuhadoc ‘to excel’:
Caja uqa ola-Ø cus-i cuha-d-on.
woman 3sg face-3sg.psr scrub-dv excel-3sg.acc-3sg.nom.remp
‘The woman cleaned her face well.’
(49) Enumerative modification expressed with a numeral:21

Q-u lecis-d-oig-a.
hit-dv two-3sg.acc-3pl.nom-todp
‘They hit it twice.’
25.3.2 Switch-Reference
Switch-reference (SR) in Amele is judged to be a local syntactic device for
monitoring the referentiality of PSA arguments between adjacent clauses as
to whether they have identical or non-identical reference (Roberts (2017).
Dependent SR verbs occur most commonly in a clause chain.22 A clause chain
comprises a string of linked clauses with only the final clause in the chain
marked for clausal operator categories, such as tense or illocutionary force.
The non-final clauses are dependent on the final clause for clausal operator
designation. The operator-dependent clauses are also typically marked with
switch-reference morphology. Amele has two basic types of ss/ds morphology.
One codes sequential events and the other codes simultaneous overlap of
events. The sequential verb is marked with either -im ‘ss.seq’ or -ec/-oc ‘ds.
seq’ followed by nom agreement morphology. The simultaneous verb is
marked with nom agreement morphology which indicates either ss.sim or
ds.sim. The ds.sim morphology is further divided into that which indicates
realis status and that which indicates irrealis status of the tense or IF
category marked on the final clause in the clause chain.23 The ds.sim verb
agrees in status value with the status value of the final clause. ss/ds clauses
are most commonly in a cosubordinate relationship with other clauses (see
Chapter 13). However, ss/ds clauses can also have a subordinate function.
25.3.2.1 Cosubordinate Switch-Reference Clauses

A typical clause chain of four linked clauses in a cosubordinate relationship
is illustrated in (50). The verbs neceb and tobocomin are marked for ds.seq
and the verb series sumudi bibiligin is marked for ds.sim.r since the tense
category of the final verb is realis status. The final verb in the chain is
belowan and this verb is marked for yesterday’s past tense and declarative
illocutionary force. Sequential events are indicated by ‘{&}’ in the LS and
simultaneous events by ‘{^}’. The clauses . . . sumudi bibiligin, neceb, tobocomin
and belowan in (50) are in a cosubordinate relationship as the tense, status
and IF categories marked on belowan have scope over all the clauses in the
chain. Sumudi bibiligin is a continuative SVC. The verb bibiligin modifies
sumudi as a continuative event and the durative marking on bi~biligin
indicates the ‘waiting’ event overlapped for a period of time with the ‘came
down’ (neceb) event.

942 JOHN R. ROBERTS
(50) Dependent switch-reference verbs:

1. Ija Malolo uqa¼na ka jic anag ono¼nu
1sg Malolo 3sg¼of car road mother there¼for
sum-ud-i bi~bil-igin
wait-3sg.acc-dv dur ~sit-1sg.nom.ds.sim.r
2. n-ec-eb
come down-ds.seq-3sg.nom
3. tob-oc-omin
ascend-ds.seq-1sg.nom
4. bel-ow-an.
go.nsg-1du.nom-yestp
〈IF DECL 〈STA REAL 〈TNS YESTP 〈ASP CONT 〈ASP DUR be-loc′ (ono, [be-loc′
(jic anag, do′ (1sg, [wait′ (1sg, become be-loc′ (jic anag, [have′ (Malolo,
ka))])〉〉 {^} do′ (3sg, [come.down′ (3sg)]) {&} do′ (1sg, [ascend′ (1sg)])
{&} do′ (1du, [move.away.from.ref.point′ (1du)])〉〉〉
‘While I waited for Malolo’s car there on the main road, he came down,
I climbed in and off we (du) went.’
Note that the final clause in a clause chain need not be finite. In (51) the
purpose clause ho bubusaleb qoc ‘to kill the pig as it runs out’ is a clause chain
with qoc ‘to kill’ as the final clause. The verb bubusaleb in the dependent
clause is marked for ds simultaneous irrealis nom agreement morphology
as the understood PSA of qoc is ‘those men’, and qoc, being infinitive, has
irrealis status.
(51) Infinitival final clause:

Dana eu ho bu~busal-eb q-oc¼nu h-oig-a.
man that pig dur ~run out-3sg.nom.ds.sim.irr hit-inf¼for come-3pl.nom-todp
‘Those men came to kill the pig as it runs out.’
A diagram of the syntax of (50) is given in Figure 25.14. The clausal

operators of tense, status and IF apply to all the dependent clauses, whereas
the core and nuclear operators apply only to a particular verb. The ss/ds
markers are treated as clause linkage markers. The rules for how the SR
system works are set out in (52) and (53). In terms of Comrie’s (1989)
different types of reference-tracking systems, SR in Amele is local and
interclausal.
(52) Protocol for selecting switch-reference controlling clause:

For a dependent clause in a cosubordinate relationship ({&} or {^}), select the
next cosubordinate clause as the controlling clause for switch-
reference marking.
(53) Amele switch-reference coreferentiality rule:

Check the pivot of the dependent clause against the pivot of the controlling
clause for coreferentiality of identity. If the referent is identical, mark ss; if
not, mark ds.
With respect to (53), there are complexities in determining referential

identity.24 For example, Amele SR is asymmetrical where there is referential

SENTENCE
CLAUSE
CLAUSE CLAUSE CLMDS CLAUSE CLMDS CLAUSE
RP CORE CORE CORE CORE
PP NUC ARGDS
NUC NUC NUC ARG NUC ARG NUC ARG
PRED PRED PRED PRED
PRO V V V V V
Ija Malolo uqana ka jic anag ononu sumudi bi~bil-igin n-ec-eb tob-oc-omin bel-owan=Ø.
NUC ASP: DUR NUC NUC NUC
NUC ASP: CONT
CORE CORE CORE CORE
CLAUSE CLAUSE CLAUSE CLAUSE
CLAUSE TNS: YESTP
CLAUSE STA: REALIS
CLAUSE IF: DECL
SENTENCE
Figure 25.14 Cosubordinate clause chain structure, example (50)
overlap between the pivot referents of the controlling clause and the pivot
referents of the dependent clause. Where the pivot referent(s) of the control-
ling clause is (are) properly included in the set of pivot referents of the
dependent clause then ss is marked. However, when the pivot referent(s) of
the dependent clause is (are) properly included in the set of pivot referent(s)
of the controlling clause then ds must be marked. For example, in (50) 1sg.
nom in [3] is properly included in 1du.nom in [4] and in this case ds must
be marked. If it was the other way around (‘we (du) climbed in and off
I went’) and 1sg.nom was the controlling referent and 1du.nom was the
dependent referent then ss would be marked.25
Stirling (1993) suggests that SR in Amele tracks events rather than the main
participant(s) in the event. As already mentioned, Amele has asymmetrical

944 JOHN R. ROBERTS
ss/ds marking when there is referential overlap between the PSA of

the controlling clause and the PSA of the marked clause. When a plural
number of participants controls a singular number of participants then ds
must be marked. In (54) a multiple (plural) event budu beladeiga with mul-
tiple participants controls a singular dependent event age cajimeig with
multiple participants. If the SR tracked events, as suggested by Stirling
(1993), then it would be expected that a move from a singular event to a
multiple event should trigger ds marking. However, ss is marked. This
shows that the Amele SR system tracks the coreferentiality of participants
across clauses (both plural in this case) rather than the coreferentiality
of events.
(54) Event quantification and SR marking:

Dana caja¼ca age caj-im-eig bud-u bel-ad-eig-a.
man woman¼add 3pl arise-ss.seq-3pl.nom disperse-dv go-distr-3pl.nom-todp
‘The people got up and dispersed in all directions (multiple events).’
Further evidence that SR in Amele tracks participants rather than events is

provided by the reciprocal verb form, as illustrated in (55). Here a pair of ds-
marked verbs functions as the predicate in the nucleus.26 The 3pl.nom
agreement at the end of the predicate refers to the whole group of women
(caja). The ds-marked verbs express the notion of one womani hitting
another womank and this being reciprocated by womank hitting womani.
Thus the ds marking refers to individual participants in the overall event of
women hitting.
(55) Reciprocal verb and SR marking:

Caja q-oc-ob q-oc-ob egi-na.
woman hit-ds.seq-3sg.nom hit-ds.seq-3sg.nom 3pl.nom-prs
〈IF DECL 〈STA REAL 〈TNS PRS do′ (3pl [caja], [seml do′ (3sg [caja]i, [hit′ (3sg [caja]i,
3sg [caja]k)]) {&} seml do′ (3sg [caja]k, [hit′ (3sg [caja]k, 3sg [caja]i)])]〉〉〉
‘The women are hitting each other.’
25.3.2.2 Subordinate Switch-Reference Clauses

Switch-reference clauses can also have a subordinate function. The
following functions are described below:
• core subordinate object clause of perception verbs

• layered cosubordination in a clause chain for modification
• ad-clausal subordination in a conditional clause
An SR clause may function as the object of a perception verb.27 In (56a)
the second argument of the perception verb perceive′ is a proposition
become arise′ (dedeman) ‘a smell had arisen’. Therefore in the syntax
dedeman waseceb occupies the DUn position in the clause between the PSA
caja eu ‘that woman’ and the verb don ‘she perceived’. In Roberts (1988a) it
is shown that overtly coordinate clauses in Amele, such as a qa ‘but’ clause,

SENTENCE
CLAUSE
RP CORE
CLAUSE CLMDS NUC ARG
RP CORE
NUC ARG
PRED PRED
N N V V
Caja eu dedeman was-ec-eb d-on =Ø.

woman that smell arise-DS.SEQ-3SG.NOM perceive-3SG.NOM.REMP
NUC NUC
CORE
CLAUSE
CORE
CLAUSE TNS: REMP
CLAUSE STA: REALIS
CLAUSE IF: DECL
SENTENCE
Figure 25.15 Core subordinate DUn clause, example (56a)
cannot be embedded within another clause. Therefore dedeman waseceb

cannot be in a coordinate relationship with don. It is core subordinate, as
illustrated in Figure 25.15. Since become arise′ (dedeman) is an accom-
plishment (ending in a result state) it is interpreted as being in a sequen-
tial relationship with the matrix predicate perceive′, that is, the smell
arose before the woman perceived it. Therefore, the verb waseceb is marked
for ds.seq.
Similarly, in (56b) the second argument of the perception predicate see′
is the proposition lie′ (3sg [ma susul] ‘the taro peelings are lying (on the
ground)’. Consequently, ma susul eu ninijen occupies the DUn position in the
clause between the PSA mala uqa ‘chicken he’ and the verb fen ‘he saw’.
Here ma susul eu ninijen is in a core subordinate relationship with fen.
Because the proposition lie′ (3sg [ma susul]) is a state it is interpreted as
occurring simultaneously with see′. The verb ninijen is therefore marked as
ds.sim.r.

946 JOHN R. ROBERTS
(56) Core subordinate DUn clause of perception verbs:

a. Caja eu dedeman was-ec-eb d-on.
woman that smell arise-ds.seq-3sg.nom perceive-3sg.nom.remp
〈IF DECL 〈STA REAL 〈TNS REMP perceive′ (3sg [caja], [become arise′ (3sg [dedeman])])〉〉〉
‘That woman perceived that a smell had arisen.’
b. Mala uqa ma susul eu ni~nij-en f-en.
chicken 3sg taro peelings that dur ~lie-3sg.nom.ds.sim.r see-3sg.nom.remp
〈IF DECL 〈STA REAL 〈TNS REMP see′ (3sg [mala], [〈ASP DUR lie′ (3sg [ma susul])〉])〉〉〉
‘Chicken saw those taro peelings lying (there).’
A switch-reference clause can function as a modifier of a following clause.

(57) is a sentence containing five clauses. Clauses [1, 3, 5] describe the
mainline events, while clauses [2, 4] provide background information.
Clause [2] says ‘when the yams ripened . . .’ and clause [4] says ‘when the
yams dried . . .’. The secondary nature of clause [2] is indicated by the ss
marking on clause [1]. In this case, clause [3] is the controlling clause for the
ss marking on clause [1]. Clause [3] is also the controlling clause for the ds
marking on clause [2]. Thus clause [2] functions as a temporal modifier to
clause [3]. Similarly, the ss marking on clause [3] is controlled by clause [5]
and clause [4], marked for ds, functions as a temporal modifier to clause [5].
There are thus layers of cosubordination, which can be schematically repre-
sented as in (57a).
(57) Layered cosubordination in a clause chain:

a. [Clause-1]SS [[Clause-2]DS Clause-3]SS [[Clause-4]DS Clause-5]
1. Ceta bahim m-i he-d-um-ei
yam store put-dv finish-3sg.acc-ss.seq-3sg.nom
2. ceta wal m-ec-eb
yam ripe put-ds.seq-3sg.nom
3. ceta eu hun-im-ei
yam that bore-ss.seq-3sg.nom
4. gulden h-oc-ob
dried come-ds.seq-3sg.nom
5. ceta bahim¼na tac-en.
yam store¼in fill-3sg.nom.remp
b. 〈IF DECL 〈STA REAL 〈TNS REMP 〈ASP COMPL [do′ (3sg i, [make′ (3sg i, ceta
bahim)]) ^ proc create′ (ceta bahim)] & ingr exist′ (ceta bahim)〉 {&}
[[ become ripened′ (ceta)] {&} [do′ (3sg i, [dig.up′ (3sg i, ceta)]]]
{&}[[become dried′ (ceta)] {&} [do′ (3sg i, Ø)] cause [become be-in′
(ceta bahim, Ø)] cause [ingr full′ (ceta bahim)]])〉〉〉
‘He finished making the yam store and, when the yams had ripened, he
dug up those yams and when they dried he put them in the yam store.’
The syntactic structure of (57) is diagrammed in Figure 25.16. The modify-

ing clauses [2] and [4] are in a cosubordinate relationship to the following
matrix clause. The matrix clause is the controlling clause of the ‘sub-
cosubordinate’ clause for the purposes of switch-reference marking.

SENTENCE
CLAUSE
CLAUSESS CLAUSESS CLAUSE
PERIPHERY CORE PERIPHERY CORE
CLAUSEDS CLAUSEDS
CORE CORE CORE
Ceta bahim mi hedumei ceta wal meceb ceta eu hunimei gulden hocob ceta bahimna tacen=Ø.
NUC NUC NUC NUC NUC
CORE ASP: COMPL CORE CORE
CLAUSE CORE CLAUSE CORE
CLAUSE TNS: REMP
CLAUSE STA: REAL
CLAUSE IF: DECL
SENTENCE
Figure 25.16 Layered cosubordination in (57)
The PSA in the sub-cosubordinate clause is different to the PSA in the matrix
clause in each case, so ds is marked.
Note that this produces an anomalous SR marking between clause [1] and
the immediately following clause [2], and between clause [3] and the imme-
diately following clause [4]. The verb in clause [1] is marked for ss when the
PSA of clause [2] is different to the PSA in clause [1]. The verb in clause [3] is
also marked for ss when the PSA of clause [4] is different to the PSA in clause
[3]. This shows that the switch-reference marking system takes account of
the layers of cosubordination. For clauses signalling the mainline events,
such as clauses [1, 3, 5] in (57) and the four clauses in (50) the SR-marking
system selects the next cosubordinate clause as the controlling clause for SR
marking. For an SR clause indicating background information, such as [2, 4]
in (57), it is interpreted being relevant to the following ‘mainline event’
expressing clause, and the SR-marking system selects that clause as the

948 JOHN R. ROBERTS
controlling clause for SR marking. A translation of (57) directly reflecting its

structure would be ‘He finished making the yam store; the yams ripened,
and then he dug up those yams; the yams dried, and then he put them in the
yam store’.
The protasis (condition clause) in a conditional sentence modifies the
apodosis (consequence clause). The condition clause in the protasis can be
a switch-reference clause. ss.seq marking for the condition clause is -if/-uf
followed by the regular ss.seq nom agreement morphology. For all other
ss/ds marking the conjunction fi ‘if’ is cliticized to the switch-reference
verb. An example of an ss condition clause is given in (58a) and of a ds
condition clause in (58b). In (58a) the ss marking actually indicates the
certainty of the consequence. In both cases, the ss/ds condition clause
can be postposed after the consequence clause and it is ad-clausal
subordination.
(58) Ad-clausal subordination in a conditional clause:

a. Qee j-i he-d-uf-eg qaga-h-ig-en.
not eat-dv finish-3sg.acc-ss.cond-2sg.nom kill-2sg.acc-1sg.nom-fut
‘If you don’t eat it all (lit. finish) I will kill you.’
b. Ija ja hud-ec-emin¼fi uqa saab man-igi-an.
1sg fire open-ds.seq-1sg.nom¼if 3sg food cook-3sg.nom-fut
‘If I light the fire she will cook the food.’
The fact that switch-reference in Amele occurs in subordinate clauses as

well as cosubordinate clauses requires a revision of the protocol for selecting
the controlling clause; (52) has therefore been modified as follows.
(59) Protocol for selecting switch-reference controlling clause:

1. For a dependent clause in a cosubordinate relationship, select the next
cosubordinate clause as the controlling clause for switch-reference marking.
2. For a dependent clause in a subordinate relationship, select the superordinate
matrix clause as the controlling clause for switch-reference marking.
25.3.2.3 Grammatical Functions and Switch-Reference

The coreferentiality rule as stated in (53) applies in most cases. However,
it does not apply in a straightforward way when the dependent or
controlling clause is an IVC. In (60a) cucuiimig ‘I feared something’ is a
state verb with two arguments. It is M-transitive. The actor (emoter)
argument is marked with 1sg.nom agreement and the undergoer (target)
argument is unspecified. Here, cucuiimig functions as the controlling clause
for ija cocobig ‘as I walked’ and the dependent clause to busali nuiga ‘I ran
away’. With respect to (53), the pivot of ija cocobig is 1sg.nom and is
coreferential with the 1sg.nom pivot of cucuiimig. Thus, cocobig is marked
ss. Then the 1sg.nom pivot of cucuiimig is coreferential with the 1sg.nom
pivot of busali nuiga and cucuiimig is marked ss. Adherence to (53) is
straightforward.

(60) a. State verb cucuiec fear′ (x, y) ‘to fear something’:

Ija co~cob-ig cucui-im-ig
1sg dur ~walk-1sg.nom.ss.sim fear-ss.seq-1sg.nom
busal-i nu-ig-a.
run away-dv go-1sg.nom-todp
‘As I walked along, I was afraid, and I fled.’
b. Impersonal verb cucuidoc feel′ (x, [afraid′]) ‘to be afraid’:
Ija co~cob-ig cucui-t-ec-eb
1sg dur ~walk-1sg.nom.ss.sim fear-1sg.acc-ds.seq-3sg.nom
busal-i nu-ig-a.
run away-dv go-1sg.nom-todp
‘As I walked along, I became afraid, and I fled.’
Now compare (60b). Here, cucuiteceb ‘I was afraid’ is an IVC.28 In this case,
the verb has a 1sg.acc ‘experiencer’ argument and a 3sg.nom ‘empty’
argument. When cucuiteceb functions as the controlling clause to ija cocobig
this verb is marked ss. Thus, the 1sg.acc argument in the IVC functions as
pivot when it is the controlling clause. However, when cucuiteceb functions as
the dependent clause to busali nuiga it is marked ds. This shows that the
pivot arguments are not being compared between these clauses. Instead, the
controller 3sg.nom argument in the IVC is compared with the pivot in the
non-IVC controlling clause. Thus, the coreferentiality rule in (53) needs to be
modified to (61). (61) applies to both regular verbs and IVCs. With a regular
verb the pivot and controller are one and the same argument so monitoring
either function does not select a different argument. However, in an IVC, the
pivot and controller are different arguments, so it makes a difference which
argument is monitored when the IVC functions as controlling clause or
dependent clause.
(61) Amele switch-reference coreferentiality rule:

Check the controller of the dependent clause against the pivot of the
controlling clause for coreferentiality of identity. If the referent is identical,
mark ss; if not, mark ds.
25.3.2.4 Pragmatic Functions of the SR System

Often in Amele text, there occur what appear to be ‘anomalous’ ds markings
where ds is indicated but the PSA of the marked clause and the controlling
clause are the same. These ds markings are, in fact, not anomalous but are
indicating a change in discourse theme. The thematic changes are primarily
in the area of time, place and possible world setting (real vs. unreal).
Thematic changes of time and place are often accompanied by temporal
and locative modifier expressions and a ds thematic change of place is most
commonly marked on a motion verb. Thematic changes in a possible-world
setting are normally a switch from the real world to an intended or pro-
posed action or vice versa, a switch from intended/proposed action to
the real world. Examples of this phenomenon are given in Roberts
(1987: 303–305, 1988b). However, these are pragmatic extensions of the SR

950 JOHN R. ROBERTS
system. At core, the switch-reference system in Amele is syntactically motiv-

ated. It is a local device for monitoring the referentiality of PSA arguments
between adjacent clauses as to whether they have identical or non-identical
reference.
References
Andersen, T. David and John R. Roberts. 1991. An exception to the hodiernal:

non-hodiernal distinction. Studies in Language 15(2): 295–299.
Comrie, Bernard. 1989. Some general properties of reference-tracking
systems. In Doug Arnold, Martin Atkinson, Jacques Durand, Claire Grover
and Louisa Sadler (eds.), Essays on Grammatical Theory and Universal Gram-
mar, 37–51. Cheltenham: Clarendon Press.
Hasegawa, Yoko. 1993. A Study of Japanese Clause Linkage: The Connective te in
Japanese. Stanford, CAS: CSLI Publications.
Imai, Shingo. 1998. Logical Structures and Case Marking in Japanese. Unpub-
lished MA project, University at Buffalo (SUNY).
Kroeger, Paul R. 2004. Analyzing Syntax. A Lexical-Functional Approach. Cam-
Latrouite, Anja and R. D. Van Valin, Jr. 2014. Event existentials in Tagalog:
A Role and Reference Grammar account. In I. Wayan Arka and N. L. K Mas
Indrawati (eds.), Argument Realisations and Related Constructions in Austrones-
ian Languages: Papers from 12-ICAL, Vol. 2, 161–174. Canberra: Asia-Pacific
Linguistics.
Lewis, M. Paul, Gary F. Simons and Charles D. Fennig (eds.). 2014. Amele. In
Ethnologue: Languages of the World (17th ed.). Dallas, TX: SIL International.
www.ethnologue.com/language/aey.
Roberts, John R. 1987. Amele. London: Croom Helm.
Roberts, John R. 1988a. Amele switch-reference and the Theory of Grammar.
Roberts, John R. 1988b. Switch-reference in Papuan languages: A syntactic or
extrasyntactic device? Australian Journal of Linguistics 8: 75–118.
Roberts, John R. 1990. Modality in Amele and other Papuan languages.
Journal of Linguistics 26: 363–401.
Roberts, John R. 1991a. Reduplication in Amele. In T. Dutton (ed.) Papers in
Papuan Linguistics, No. 1, 115–146. Canberra: The Australian National
University.
Roberts, John R. 1991b. A study of the dialects of Amele. Language and
Linguistics in Melanesia 22(1–2): 67–126.
Roberts, John R. 1991c. Orthography reform in Amele: Part One. Notes on
Literacy 17(4): 1–20.
Roberts, John R. 1992. Orthography reform in Amele: Part Two. Notes on
Literacy 18(1): 1–32.

Roberts, John R. 1994. The category ‘irrealis’ in Papuan medial verbs. Notes on
Roberts, John R. 1996. A Government and Binding analysis of the verb in
Amele. Language and Linguistics in Melanesia 27(1): 1–66.
Roberts, John R. 1997a. Switch-reference in Papua New Guinea:
A preliminary survey. In A. Pawley (ed.), Papers in Papuan Linguistics,
No. 3, 101–241. Canberra: The Australian National University.
Roberts, John R. 1997b. The syntax of discourse structure. Notes on Translation
11(2): 15–34.
Roberts, John R. 1997c. GIVE in Amele. In John Newman (ed.), The Linguistics of
Giving, 1–33. Amsterdam: John Benjamins. www.sil.org/resources/archives/
22980.
Roberts, John R. 2001. Impersonal constructions in Amele. In Alexandra Y.
Aikenvald, R. M. W. Dixon and Masayuki Onishi (eds.), Non-Canonical
Marking of Subjects and Objects (Typological Studies in Language 46),
Roberts, John R. 2012. Serial verbs in English: An RRG analysis of catenative
verb constructions. Functions of Language 19(2): 201–234.
Roberts, John R. 2015a. Distributives in Amele. A Role and Reference Gram-
mar analysis. SIL Electronic Workpapers 2015–001. www.sil.org/resources/
publications/entry/60478.
Roberts, John R. 2015b. Inalienable possession in Amele: A Role and Refer-
ence Grammar account. SIL Electronic Workpapers 2015–002. www.sil.org/
resources/publications/entry/60480.
Roberts, John R. 2017. The typology of switch-reference. In Alexandra Y.
Aikhenvald and R. M. W. Dixon (eds.), Handbook of Linguistics Typology,
Stirling, Lesley. 1993. Switch-Reference and Discourse Representation. Cambridge:
Talmy, Leonard. 2007. Lexical typologies. In Timothy Shopen (ed.), Language
Typology and Syntactic Description, 66–168. Cambridge: Cambridge University
Press.
Toratani, Kiyoko. 2002. The Morphosyntax and the Logical Structures of Compound
Verbs in Japanese. Unpublished PhD dissertation, University at Buffalo (SUNY).
Z’graggen, John A. 1975. The Languages of the Madang District, Papua New Guinea
(Pacific Linguistics B-41). Canberra: The Australian National University.
Notes
1 See Roberts (1987) for a full linguistic description of the Amele language
from a typological perspective. The dialect of Amele represented in this
study is the Haija dialect (Roberts 1991b).

952 JOHN R. ROBERTS
2 See Roberts (1997b) for the wider consequences of this feature.

3 See Roberts (1996) for an account of PSA agreement morphology and
Roberts (2001) for an account of DCA agreement morphology.
4 Non-finite includes infinitive, serial and switch-reference verbs, none of
which are marked for tense.
5 In the Amele orthography c is [Ɂ] and q is [g͡ b]. See Roberts (1991c, 1992)
for a full account of Amele orthography issues.
6 The form of acc morphology with qoc ‘to hit’ is irregular.
7 See Roberts (1997c) for a full account of ‘give’ in Amele.
8 Kin terms in Amele are divided morphologically into kin orientation
terms (have.as.orientation.kin′ (x, y)) and kin procreation terms (have.as.
procreation.kin′ (x, y)) (Roberts 2015b).
9 The verb belec ‘to go’ requires a non-singular PSA.
10 Andersen and Roberts (1991) find that in one dialect of Amele the today’s
past tense does not occur. There is only present tense and yesterday’s
past tense.
11 Prospective tense is expressed by a reduced form of the future tense
suffix, -e(n)/-a(n), together with the auxiliary posture verb bil-ec ‘to sit’.
12 They have the same pronunciation and same written form as the corres-
ponding accusative agreement markers.
13 See Roberts (1991a).
14 See Roberts (2015b) for a full account of alienable and inalienable posses-
sion in Amele.
15 Kinship terms can also express the plurality of the possessed, e.g. ate-ni-el
[daughter-1sg.psr-pl.psd] ‘my daughters’.
16 See Roberts (2012) for an application of these defining characteristics to
catenative verbs in English.
17 Note that Kroeger (2004) does not recognize that operators such as tense,
aspect, modality, negation have scope over different layers of clause
structure. In RRG, instead, aspect is a nuclear operator and, therefore,
it is possible in this framework to capture the fact that individual verbs
in an SVC can be marked for different types of aspect.
18 This is analogous to the connective -te in Japanese (Hasegawa 1993).
19 There are some serial verb constructions in which this is not the
case, though.
20 Bare ‘&’ and ‘^’ are used within the LS of verbs to signal sequential and
simultaneous subparts of a single event, whereas ‘{&}’ and ‘{^}’ indicate
sequential and simultaneous events in complex sentences.
21 Lecis-doc is the ordinal form of this numeral. See Section 25.2.4.1.
22 Switch-reference in clause chains is common in PNG languages
(Roberts 1997a).
23 See the status operator in Section 25.2.3.2 for details of which categories
are realis and which are irrealis.
24 Another complexity is how the coreferentiality rule (53) applies to IVCs.
See grammatical functions and SR in Section 25.3.2.

25 For a fuller account of referential overlap in Amele and other Papuan

languages, see Roberts (2017).
26 There are two pieces of evidence that qocob qocob functions as the predi-
cate in the nucleus. First, qocob qocob may be substituted for the verb
stem in the infinitive form. Cf. q-oc ‘to hit’ with qocob qocob ec ‘to hit each
other’. Second, durative aspect (C)V~ reduplication on the verb is marked
on the matrix nom agreement in the reciprocal verb form, e.g. q-oc-ob q-oc-
ob e~egin [hit-ds.seq-3sg.nom hit-ds.seq-3sg.nom dur ~3pl.nom.ds.
sim.r] ‘while they hit each other’. Cf. (29) where durative (C)V~ redupli-
cation is marked on the nom agreement within the verb word.
27 Contrary to what was stated in Roberts (1988a), nominal clauses in Amele
can be marked for ss/ds.
28 The semantic difference between cucuiec and cucuidoc is that with cucuiec
the emotion of fear is self-generated, whereas with cucuidoc the fear is a
response to an unspecified outside stimulus. Also, with the M-transitive
verb, the target argument can be marked on the verb by acc agreement,
cucui-ad-ec ‘to fear them’, whereas with the IVC the target argument has
to be expressed in a PP, age¼nu cucui-d-oc ‘to be afraid of them’.

26
Case and Voice in Amis
(Austronesian, Taiwan)
Joy J. Wu
Abbreviations
The abbreviations and symbols used in the glosses in general follow the
Leipzig glossing rules (www.eva.mpg.de/lingua/resources/glossing-rules.
php), with the following additions:
ABLT abilitative NFIN non-finite

ASP aspect marker PPN personal proper noun
AV actor voice PrCS pre-core slot
CN common noun PrDP pre-detached position
FAC factual mood PREP preposition
IA instrumental applicative RED reduplication
LA locative applicative UV undergoer voice
LNK linker VOL volitative mood
NEUT neutral voice
26.1 Introduction
This chapter presents a sketch grammar of Amis with a focus on the discus-
sion of the phenomena related to its case marking and voice. Amis is
currently spoken by 210,839 speakers in Taiwan.1 It is classified as one of
the Eastern Formosan branch languages (Blust 1999) and is reported to
consist of five dialects (Tsuchida 1988). The following discussion is based
on the data collected from the Central dialect.
26.2 Basic Clause Patterns
Amis is a predicate-initial language. Three types of predicates can be identi-

fied: verbal, nominal and prepositional. One-place, two-place and three-place
verbal predicates and their logical structures are provided in (1):2

Case and Voice in Amis 955
(1) Amis verbal predicates
a. R<om>akat kako i lalan.

<neut> walk 1s.nom prep road
‘I am walking on the road.’
a′. be-on′ (lalan, [do′ (kako, [walk′ (kako)])
b. Mi-palo Ø-ci Sawmah ci Mayaw-an.
av-beat nom-ppn Sawmah ppn Mayaw-dat
‘Sawmah is beating Mayaw.’ Or ‘Sawmah is going to beat Mayaw.’
b′. do′ (Sawmah, [go′ (Sawmah)]) & INGR be-at′ (y, Sawmah) PURP
do′ (Sawmah, [beat′ (Sawmah, Mayaw)])
b′′. do′ (Sawmah, [beat′ (Sawmah, Mayaw)])
c. Ma-palo n-i Sawmah Ø-ci Mayaw.
uv-beat gen-ppn Sawmah nom-ppn Mayaw
‘Mayaw was beaten by Sawmah.’
c.′ do′ (Sawmah, [beat′ (Sawmah, Mayaw)]) & BECOME beaten′ (Mayaw)
d. Pa-nanom kako t-o wawa t-o sayta.
caus-water 1s.nom dat-cn child dat-cn soda
‘I gave the child soda to drink.’ (Causative, AV)
d′. do′ (kako, Ø ) CAUSE BECOME have.water′ (wawa, sayta)
As shown in (1), verbal predicates are usually affixed with a voice marker
(e.g. mi-), followed by various numbers of arguments that are marked by a
case marker, be it a nominative case (for the privileged syntactic argument
(PSA)), a genitive case (for a non-PSA actor or a possessor), or a dative case (for
an oblique core argument or an adjunct). A locative argument or adjunct is
marked by the preposition i.
Nominal predicates and prepositional predicates are exemplified respect-
ively in (2a) and (2b).
(2) Amis nominal and prepositional predicates
a. O singsi cingra.
cn teacher 3s.nom
‘He is a teacher.’
a′. be′ (cingra, [teacher′])
b. I loma’ Ø-ci Mayaw.
prep house nom-ppn Mayaw
‘Mayaw is at home.’
b′. be-at′ (loma’, Mayaw)
Unlike verbal predicates, nominal predicates are preceded by a noun class

marker3 and prepositional predicates are initiated by a preposition.
Special clause types, such as displacement constructions (e.g. 3a) and topic
constructions (e.g. 3b), are given below.4

956 JOY J. WU
(3) Amis displacement construction and topic construction
a. O fafahian a singsi k-o ka-olah-an¼ako

cn woman lnk teacher nom-cn nfin-like-la¼1s.gen
‘It is female teachers that I like (better).’
b. O sa-pi-senger t-o felac i, nga’ay ma-lalo
cn ia-nfin-soak dat-cn rice top good.for uv-soft
o sanay.
cn like.this
‘Regarding the reason to soak the rice, it is for softening (it).’
Both of the constructions contain a displaced element that appears before the
main predicate, but there is no pause before the element in the displacement
construction and the rest of the clause while there is one in the topic construc-
tion, which means that the displaced elements in (3a) and (3b) are placed
under the pre-core slot (PrCS) and pre-detached position (PrDP) respectively.
26.3 Lexical Categories
The discussion of lexical categories includes the roots and derived words in
Amis. As pointed out by Wang (1976), all the root forms in Amis are syntactic-
ally nominal.5 In other words, to serve as a verb, root forms, even for roots
with inherent verbal meaning, have to be either affixed with a verbal prefix
(usually a voice marker6) or placed in the predicate position. The former
strategy is more commonly found and examples can be seen in (1); the latter
is only found with some state predicates such as miming ‘small’ and tosa ‘two’.
Derived verbs with no verbal affixes are termed unaffixed verbs in Wu (2006a).
As root forms are syntactically nominal, they can readily appear after a case
marker or a preposition (e.g. (4a)) to manifest arguments and adjuncts. How-
ever, the derived verbs (e.g. root form plus a voice marker) have to be nominal-
ized7 before they can show up in a nominal position (e.g. (4b–c))8. See Wu
(2006a: 70) for some common nominalizing or deverbal strategies in Amis.
(4) Amis nouns and deverbal nouns
a. Na’on-en k-o rakat!

mind-uv nom-cn walk
‘Good-bye.’ Lit. ‘Mind your walk!’
b. *Na’on-en k-o r<om>akat!
mind-uv nom-cn <neut>walk
c. Na’on-en k-o r<om>akat-ay!
mind-uv nom-cn <neut>walk-fac
‘Mind the one who is walking!’
The manifestation of predicates was illustrated in Section 26.2; a verbal

predicate is expressed by a derived verb (e.g. (1)), a nominal predicate is
composed of a noun (derived or base-generated) preceded by a noun class

marker (e.g. (2a)), and a prepositional predicate is designated by a prepos-

ition plus a noun (e.g. (2b)).
26.4 Macrorole Selection and Argument-Marking
This section discusses the macrorole selection of Amis predicates with

different numbers of core arguments and how arguments and non-
arguments are marked.
26.4.1 Macrorole Selection of Verbs with Various Numbers

of Core Arguments
Amis predicates with zero core arguments are found in meteorological or
phenomenal verbs such as ma-’orad ‘rain’, and sienaw ‘cold (in terms of
weather)’. These verbs can appear by themselves without any co-occurring
argument, as illustrated in (5a). As there is no core argument, there is no
macrorole for such verbs, and thus, they are macrorole-atransitive (M-atran-
sitive). However, such verbs can also take one core argument, as illustrated
in (5b) and become M-intransitive. Their different logical structures are
given in (5a′) and (5b′) respectively.
(5) Amis meteorological or phenomenal verbs
a. Ma-’orad anini.
neut-rain now
‘It is raining today.’
a′. rain′ (Ø)
b. Ma-’orad k-o kakarayan.
neut-rain nom-cn sky
‘The sky is raining.’
b′. rain′ (kakarayan)
Regarding the assignment of the macrorole for verbs with one core argu-
ment (i.e. syntactically intransitive or S-intransitive verbs), the presence or
absence of do′ in their logical structure (LS) makes a crucial difference; the
presence of do′ makes the only core argument an actor while the absence of
do′ makes it an undergoer. In other words, the single argument for intransi-
tive activity verbs such as t<om>ireng ‘stand’ and intransitive state verbs
such as ma-toni’ ‘soft’ will not be assigned the same macrorole:
(6) Amis verbs with one core argument
a. T<om>ireng cingra.
<neut>body 3s.nom
‘He is standing.’
b. Ma-toni’ k-o-ni a titi.
neut-soft nom-cn-this lnk meat
‘This meat is soft.’

958 JOY J. WU
Although the only argument of the two verbs in (6) ia marked by the same
case (i.e. the nominative case), the macrorole of t<om>ireng would be actor
while that of ma-toni’ would be undergoer. This is where the Role and
Reference Grammar (RRG) analysis differs from other studies concerning
this issue in Amis (e.g. Liu 1999, Liu 2003, to name just a few). In such
research, the single arguments of S-intransitive verbs seem to be treated as
the same type and the verbs are labelled as actor-focus or actor-voice verbs.
This analysis will be further discussed in the last section of this chapter.
When there are two core arguments in the LS of an Amis predicate, the
macrorole assignment follows the Actor-Undergoer Hierarchy (AUH) and the
default macrorole assignment principles. However, it is also possible for
such verbs to have only one macrorole. As illustrated in (1b) and (1c), two-
place verbs can appear with two case-marking patterns: the nominative-
dative case frame in the actor voice (AV) sentences and the genitive-
nominative pattern in the undergoer voice (UV) sentences. While the two-
place UV verb follows the AUH and has both an actor and an undergoer, the
second argument in their AV counterparts is a non-macrorole core argu-
ment, as argued in Wu (2006a). In other words, the two-place AV verbs are M-
intransitive. The non-macrorole status of the second argument results either
from the linking phase from the argument position in the LS to macrorole
or from the voice operation that deprives a core argument of its macrorole
status during linking from macrorole to syntactic function. The first case is
found when the second argument of an activity verb is non-referential, as
exemplified in (7a). The second one is illustrated in (7b).
(7) Amis AV sentences with a two-place predicate
a. Mi-nanom cingra (t-o nanom).

av-water 3s.nom dat-cn water
‘He is drinking water.’ Or ‘He is going to drink water.’
b. Mi-nanom cingra t-o-ra sayta.
av-water 3s.nom dat-cn-that soda
‘He is drinking that soda.’ Or ‘He is going to drink that soda.’
The principal difference between a macrorole second argument (normally a

patient) and a non-macrorole one lies in the possibility for this argument to
be promoted by the applicative construction; only a non-macrorole argu-
ment (or an adjunct) is eligible to appear in such constructions. See Wu
(2006a) for the discussion in detail.
In addition to the two-place AV verbs, examples illustrating the mismatch
between syntactic transitivity and macrorole transitivity is also seen in three-
place predicates. To begin with, the AV construction of a three-place predicate
is also deemed M-intransitive. Both of its non-actor arguments are marked by
the dative case and, most important of all, both of them can be the promoted
argument in the applicative construction, which is a feature for a non-
macrorole argument in the AV construction. Examples are given in (8).

(8) Amis three-place AV predicate and the applicative constructions
a. Pa-nanom cingra ci Aki-an t-o-ra sayta.

caus-water 3s.nom ppn Aki-dat dat-cn-that soda
‘He gave Aki that soda (to drink).’ (Causative, AV)
b. Cima k-o pa-nanom-an nira t-o-ra sayta?
who.nom nom-cn caus-water-la 3s.gen dat-cn-that soda
‘Who did he give that soda to drink?’ (causee/recipient as the undergoer)
c. O maan k-o pa-nanom-an nira ci Aki-an?
cn what nom-cn caus-water-la 3s.gen ppn Aki-dat
‘What did he give to Aki to drink?’ (causand/theme as the undergoer)
As shown in (8), both the recipient argument and the theme argument can
be the PSA,9 in particular, the syntactic pivot, of the applicative three-place
verb pa-nanom-an ‘give water’. This is a property for a non-macrorole (NMR)
argument. In other words, neither of the two non-actor arguments in AV pa-
sentences such as (8a) are macroroles.
While the UV three-place predicates are M-transitive, just like the UV two-
place UV predicates, the mismatch between S-transitivity and M-transitivity is
still found as there can be at most two macroroles in RRG, while there are three
core arguments in a three-place predicate. This means there is competition for
macrorole status between the two groups of potential undergoer participants,
namely, theme/patient and recipient/beneficiary/source/goal. Hence, two prin-
ciples of undergoer selection (i.e. Principle A: choosing the lowest-ranking
argument in LS and Principle B: choosing the second-highest-ranking argu-
ment in LS) have been proposed in RRG under the AUH. As reported in Guerrero
Valenzuela and Van Valin (2004), languages tend to exhibit a mixed type that
needs both principles to account for their undergoer selection. Amis is also
such a language. Wu (2006a, b) has shown that Amis three-place predicates vary
in their selection of the undergoer, though Principle B seems to be more
commonly employed. The evidence lies in the possibility of having either the
theme/patient or the recipient/beneficiary/source/goal argument as the PSA in
the UV constructions, and the latter group seems to be favoured by more three-
place predicates.10 Some examples are given below. As one can see, verbs like
pa-caliw ‘lend’ can only have the theme argument as the PSA in their UV
construction, while verbs like pa-nanom ‘give water’ only allow the recipient
argument as the PSA in their UV counterpart.
(9) Some UV three-place predicates
a. Ma-pa-cakay n-i Aki k-o foting ci Ofad-an.

uv-caus-buy gen-ppn Aki nom-cn fish ppn Ofad-dat
‘Aki sold (other people’s) fish to Ofad.’
a′. [do′ (Aki, Ø)] CAUSE [[do′ (Ofad, Ø)] CAUSE [BECOME NOT have′
(Aki, foting) & BECOME have′ (Ofad, foting)]]
b. *Ma-pa-cakay n-i Aki t-o foting Ø-ci Ofad.
uv-caus-buy gen-ppn Aki dat-cn fish nom-ppn Ofad

960 JOY J. WU
c. Ma-pa-nanom¼to n-i ina t-o sayta Ø-ci

uv-caus-water¼asp gen-ppn mother dat-cn soda nom-ppn
mama.
father
‘Mother gave Father soda to drink.’
c′. [do′ (ina, Ø)] CAUSE [BECOME have.water′ (mama, sayta)]11
d. *Ma-pa-nanom¼to n-i ina ci mama-an k-o
uv-caus-water¼asp gen-ppn mother ppn father-dat nom-cn
sayta.
soda
After briefly reviewing how the macroroles are selected for verbs with
various numbers of core arguments, now let us see how these arguments
and non-arguments are marked in Amis.
26.4.2 Argument-Marking
Arguments and non-arguments are marked in Amis through a tri-case
system (nominative, genitive, dative) and the preposition i. The nominative
case marks the PSA, or the so-called grammatical subject. The genitive case
can mark a possessor or an actor in a non-AV sentence (e.g. (1c)), which
makes it the equivalent marker of an ergative case. The dative case serves a
wide range of functions. It can label a NMR core argument, as seen in the
above examples (e.g. (1b)). It can also appear before an oblique core argu-
ment or an adjunct. Examples follow.12
(10) Amis dative case marking an oblique core argument or an adjunct
a. Ma-pa-nanom¼to n-o wawa k-o kolong

uv-caus-water¼asp gen-cn child nom-cn water.buffalo
t-o-ya nanom.
dat-cn-that water
‘The child has already fed the water buffalo that water.’ (Theme)
a′. Ya nanomi ma-pa-nanom¼to n-o wawa k-o
that water uv-caus-water¼asp gen-cn child nom-cn
kolong ____i.
water.buffalo
‘That water the child has already fed the water buffalo.’
b. Ma-ota’ kako t-o sanek n-o tosiya.
neut-vomit 1s.nom dat-cn smell gen-cn car
‘I feel sick from the smell of cars.’ (Reason)
c. Ma-tayal kako t-o romi’ami’ad.
neut-work 1s.nom dat-cn every.day
‘I work every day.’ (Time)
d. Cenger-en¼ako k-o kiladom t-o kohting-ay.
color-uv¼1s.gen nom-cn cloth dat-cn black-fac
‘I am going to colour the cloth with the black colour.’ (Instrument)
The dative case marks an oblique core argument in (10a) and various kinds of
adjuncts in (10b–d). The oblique status of the theme argument in (10a) is

indicated by its displaced version in (10a′), in which it is simply displaced to the

sentence-initial position without affecting the rest of the sentence. This is the
feature of displacing an oblique reference phrase (RP) in Amis; the displace-
ment of a direct core RP will result in a nominal structure like an equational
sentence, as will be seen in the later discussion. An Amis RP that appears clause-
initially, be it a displaced one or not, does not bear any case marking. That is
why the demonstrative pronoun before nanom ‘water’ has no case in (10a′).
Although both NMR core arguments and adjuncts can be marked by the
dative case, there is a crucial difference between them, which lies in the
mechanisms to promote their status to PSA. The core arguments can be pro-
moted via plain undergoer voice constructions and applicative constructions:
(11) The strategies for promoting an NMR core argument
a. Ma-nanom n-i Aki k-o-ra sayta.

uv-water gen-ppn Aki nom-cn-that soda
‘Aki drank that soda.’ (Plain UV)
b. Mi-nanom-an n-i Aki k-o sayta.
la-water-la gen-ppn Aki nom-cn soda
‘Aki drank the soda.’ (Locative applicative UV)
‘What Aki drank is the soda.’ (Locative applicative UV)
c. Olah-en namo ∅-ci Panay.
like-uv 2p.gen nom-ppn Panay
‘You have to love Panay.’ (Plain UV)
d. Ka-olah-an¼ako ∅-ci Panay.
nfin-like-la¼1s.gen nom-ppn Panay
‘Panay is the one I like (most).’ (Locative applicative UV)
The data in (11) illustrates the possibilities for enhancing the status of the
second argument of pred′ (the one marked by the dative case) in mi-nanom
and ma-olah to become a PSA (i.e. undergoer of a UV verb). Both the plain UV
constructions (e.g. (11a) and (11c)) and the applicative UV constructions (e.g.
(11b) and (11d)) are applicable here. Note that the number of core arguments
in the two predicates has remained the same in the plain UV constructions
and the applicative UV constructions.13
Now consider a different case exemplified in (12).
(12) The strategies for promoting an adjunct
a. Ma-patay k-o ’oner t-o sapaiyo n-o ’edo.

neut-dead nom-cn snake dat-cn medicine gen-cn mouse
‘Snakes may die from the poison for killing mice.’
a′. (BECOME) dead′ (oner)
b. Sa-pi-patay n-o mato’asay t-o ’oner k-o
ia-nfin-dead gen-cn old.man dat-cn snake nom-cn
sapaiyo n-o ’edo.
medicine gen-cn mouse
‘The old man killed the snake with the poison for killing mice.’
(Instrument applicative, UV)

962 JOY J. WU
b′. [do′ (mato’asay, [use′ (mato’asay, sapaiyo no ’edo)])] CAUSE [[do′ (sapaiyo no
’edo, Ø) ]CAUSE [BECOME dead′ (’oner)]]
c. Ma-ota’ kako t-o sanek n-o tosiya.
neut-vomit 1s.nom dat-cn smell gen-cn car.
‘I feel like vomiting from the smell of cars.’
c′. do′ (kako, [vomit′ (kako, (y))])
d. Sa-ka-ota’ ako k-o sanek n-o tosiya.
ia-nfin-vomit 1s.gen nom-cn smell gen-cn car
‘The smell of the car is the reason why I vomit.’ (Instrumental applicative, UV)
d′. because.of′ (sanek no tosiya, [do′ (kako, [vomit′ (kako, (y))])
As shown in (12), for a reason/indirect cause adjunct RP marked by the dative

case to become a PSA, only the applicative construction can be used. Note that
the number of core arguments will change when the applicative constructions
are employed, as can be seen from the comparison of the number of arguments in
the LS of the non-applicative verb (e.g. (12c′)) and the applicative one (e.g. (12d′)).
Another way to make the adjunct in (12a) and (12c) a PSA, in addition to
using the applicative UV construction, is to make the adjunct the actor of an
AV construction, as illustrated in (13):
(13) The strategies for promoting an adjunct
a. Mi-patay k-o sapaiyo n-o ’edo t-o ’oner.

av-dead nom-cn medicine gen-cn mouse dat-cn snake
‘The poison for killing mice may kill a snake as well.’
a′. [do′ (sapaiyo no ’edo, Ø)] CAUSE [BECOME dead′ (’oner)]
b. Mi-ota’ t-o tamdaw k-o sanek n-o tosiya
av-vomit dat-cn person nom-cn smell gen-cn car
‘The smell of cars makes people vomit.’
b′. [do′ (sanek no tosiya, Ø)] CAUSE [BECOME vomit′ (tamdaw)]
The reason/indirect cause adjunct RPs in (12a) and (12c) now become the actors
in (13a) and (13b), respectively. As indicated in the logical structures of mi-patay
in (13a) and mi-ota’ in (13b), the predicates have become causativized, and there
is an effector added to the core of the predicates.14 In other words, the number
of core arguments has also been changed. The addition of a core argument is
not found in the examples in (11); when the to RPs of mi-nanom and ma-olah
become PSAs in the plain or applicative UV constructions, there is no argu-
ment addition involved. Hence, the to RP of ma-patay in (12a) and ma-ota’ in
(12c) should be analysed differently from the to RPs in mi-nanom in and ma-olah;
the former are adjuncts while the latter are NMR core arguments.
The last difference between an NMR direct core argument and an
adjunct (or even an oblique core argument) is that the status of the
former can always be adjusted through the plain voice operation; how-
ever, for the latter, it is not always possible. In other words, some to RPs
can only be promoted by means of the applicative constructions. For
example, the plain voice construction is quite unlikely to be employed to
promote the adjunct manifesting temporal expression in (10c), although the

applicative form ka-tayal-an ‘place or time for working’ can be used. For some
adjuncts that are more likely to be construed as effectors (e.g. an indirect cause
like sanek no tosiya ‘smell of the car’ in (12c), they may be promoted to become
an actor in AV and UV constructions, as we have seen in (13).
Based on the discussion so far, the case assignment rules in Amis can be
summarized as (14):15
(14) Case assignment rules in Amis
a. Assign nominative case to the lowest macrorole argument in terms of the

PSA Selection Hierarchy.
b. Assign genitive case to the other macrorole argument.
c. Assign dative case to other direct core argument(s).
The examples in (15) illustrate how the rules in (14) are applied in Amis.
(15) The application of case assignment rules in Amis
a. Ma-olah kako ci Panay-an

av-like 1s.nom ppn Panay-dat
‘I like Panay,’
a′. like′ (kako, Panay)
(Rule(s) applied: (14a) and (14c))
b. Ma-ka-olah¼ako Ø-ci Panay.
uv-nfin-like¼1s.gen nom-ppn Panay
‘I love Panay (secretly).’ or ‘Panay was loved by me.’
b′. like′ (ako, Panay). . ..BECOME like′ (ako, Panay)
(Rule(s) applied: (14a) and (14b))
c. Ma-stol kako t-o fekeroh.
neut-stumble 1s.nom dat-cn rock
‘I stumbled over on the rock.’
c′. stumble′ (kako)
(Rule(s) applied: (14a))
d. Ma-stol n-o fekeroh kako.
uv-stumble gen-cn rock 1s.nom
‘The rock rolled to me and made me stumble.’
d′. [do′ (fekeroh, Ø)] CAUSE [BECOME stumble′ (kako)]
e. Ma-rohem¼to k-o-ra pawli.
neut-ripe¼asp nom-cn-that banana
‘The banana has become ripe.’
e′. (INGR/BECOME) ripe′ (pawli)
f. Pa-si-fana’ k-o singsi t-o wawa t-o
caus-have-knowledge nom-cn teacher dat-cn child dat-cn
n-o Amis.
gen-cn Amis
‘The teacher is going to teach the children Amis.’ (Causative, AV)
f′. [do′ (singsi, Ø)] CAUSE [BECOME have.knowledge′ (wawa, no ’Amis)]

964 JOY J. WU
In addition to a set of case markers, there is also a preposition i in Amis.

This preposition mainly marks arguments with a locative feature (i.e. x in
be-loc′ (x, y) or pred-loc′ (x, y)). In addition, it also marks the first argument
of the existential verbs ira or awa (i.e. (NOT) exist′ ([pred′ (x, y)])) and possibly
the first argument in the embedded logical structure BECOME/INGR pred′ (y,
z). Some examples are given in (16):
(16) The functions of the preposition i
a. Maro’ kako i Taypak.

live 1s.nom prep Taipei
‘I live in Taipei.’ (Neutral voice)
a′. live.in′ (Taypak, kako)
b. Ira k-o kawas i loma’ nira.
exist nom-cn ghost prep house 3s.gen
‘There is ghost in his house.’ (Neutral voice)
b′. exist′ ([be-in′ (loma’ nira, kawas)])
c. Pa-nengneng kako t-o-ni~ni t-o/i wawa.
caus-see 1s.nom dat-cn-this~red dat-cn /prep child
‘I showed the child this.’ Or, ‘I showed this to the child.’ (Causative, AV)
c′. [do′ (kako, Ø)] CAUSE [BECOME see′ (wawa, tonini)]
However, as illustrated in (16), while the first argument of the embedded

BECOME/INGR pred′ might be marked in more than one way (e.g. by a dative
case or a preposition), the preposition is the only choice for the first argu-
ment of pred-loc′ (x, y). Moreover, while the first argument of the embedded
BECOME/INGR pred′ can be a possible undergoer and hence a PSA in the UV
construction, it is impossible for the first argument of pred-loc′ to be an
undergoer, let alone a PSA. This is illustrated by the following contrast
between pa-nanom ‘cause to have water’ and pa-teli ‘put’ in (17):
(17) Comparison of the first argument of BECOME/INGR pred′ and pred-loc′
a. Pa-nanom-en k-o sayta t-o nanom!

caus-water-uv nom-cn soda dat-cn water
‘Add water to the soda!’
a′. DO (x, [do′ (x, Ø)] CAUSE [BECOME have.water′ (sayta, nanom)]
b. Pa-teli kako t-o konga i langa.
caus-put 1s.nom dat-cn sweet.potato prep basket
‘I put the sweet potatoes in the basket.’ (Causative, AV)
b′. [do′ (kako, Ø)] CAUSE [BECOME be-loc′ (langa, konga)]
c. Ma-pa-teli’¼ako k-o konga i langa.
uv-caus-put¼1s.gen nom-cn sweet.potato prep basket
‘I put the sweet potato in the basket.’
c′. *Ma-pa-tli’¼ako t-o konga k-o langa.
uv-caus-put¼1s.gen dat-cn sweet.potato nom-cn basket
As indicated in (17), the first argument of the embedded be-loc′ (e.g. langa
‘basket’)) cannot be an undergoer in the UV construction. This follows
from the claim in RRG that the first argument of be-loc′ or pred-loc′

cannot be a macrorole; in other words, two-place locative predicates are

always M-intransitive. The following preposition-assignment rules are postulated
for Amis:
(18) Preposition-assignment rules for Amis
Assign the preposition i to the first argument of . . .pred′ (x, y). . . if it is a non-
macrorole argument:
(i) obligatory if pred′ (x, y) ¼ pred-loc′ (x, y), x ¼ common noun
(ii) optional if pred′ (x, y) ¼ pred-loc′ (x, y), x ¼ personal proper noun
(iii) optional if pred′ (x, y), pred′ ¼ cognition, possession, and perception
26.5 Grammatical Relations and the Voice System
This section discusses grammatical relations and the voice system in Amis.
According to Wu (2006a), Amis only employs a PSA in the formation of
relative clauses, a nominal type of displacement constructions and wh-ques-
tions. Constructions that might need a grammatical relation in other lan-
guages, such as control constructions and reflexivization, can be taken care
of by semantic roles in Amis. Hence there are no grammatical relations
found in such constructions.
As for the voice system, Amis distinguishes two voices: actor and under-
goer, and the latter can be further divided into the plain (or non-applicative)
UV and the applicative UV. The functions of each voice will be elaborated in
the following sub-sections.
26.5.1 Grammatical Relations

A relative clause (RC) in Amis is formed by gapping an RP from the modify-
ing clause. The gapped RP is a pivot as it is omitted in the clause. This gapped
RP is coreferential with the modified noun that follows the RC. To serve as a
head of an RC, its coreferential gapped RP has to be the actor of an AV verb,
the undergoer of a plain UV verb, or an applied argument of an applied UV
verb in the RC. If the gapped RP does not belong to any of these types, the
sentence will be rendered ungrammatical. This is exemplified in (19), in
which the gapped RP is indicated by ‘__’ in the RC.16 Hence, there is
restricted neutralization of semantic roles on the pivot of an RC in Amis.
(19) Amis relative clauses
a. Pivot: Actor of AV verb

Ma-patay¼to k-o-ya mi-kalat-ay ____i ci
neut-dead¼asp nom-cn-that av-bite-fac ppn
Aki-an a wacoi.
Aki-dat lnk dog
‘That dog that bit Aki is dead.’

966 JOY J. WU
a′. Pivot: NMR direct core argument of AV verb

*Ma-patay¼to k-o-ya mi-kalat-ay k-o waco ____i
neut-dead¼asp nom-cn-that av-bite-fac nom-cn dog
a tamdawi.
lnk person
‘That person that the dog bit is dead.’
b. Pivot: Undergoer of UV verb
Tati’ih k-o-ya ma-kaen-ay n-i Aki ____i a
bad nom-cn-that uv-eat-fac gen-ppn Aki lnk
talii.
taro
‘That taro that Aki ate was bad.’
b′. Pivot: (Patient) undergoer of applied UV verb
Tati’ih k-o-ya mi-kaen-an n-i Aki ____i a
bad nom-cn-that la-eat-la gen-ppn Aki lnk
talii.
taro
‘That taro that Aki ate was bad.’
c. Pivot: Actor of UV verb
*Ma-soso k-o-ya ma-kaen-ay ____i k-o tali
neut-fat nom-cn-that uv-eat-fac nom-cn taro
a tamdaw.
lnk person
‘The person that ate the taro was fat.’
d. Pivot: (Instrument) undergoer of applied UV verb
Ma-pitek¼ako k-o sa-pi-cikcik n-i Aki
uv-break¼1s.gen nom-cn ia-nfin-cut gen-ppn Aki
t-o dateng ____i a po’oti.
dat-cn vegetable lnk knife
‘I broke the knife with which Aki cut the vegetable.’
e. Pivot: (Locative) undergoer of applied UV verb
Tayra Ø -ci Panay mi-ladom i
go nom-ppn Panay neut-fetch.water prep
pi-ladom-an n-i Aki ____i a tefoni.
nfin-fetch.water-la gen-ppn Aki lnk well
‘Panay went to fetch water at the well where Aki fetched water.’
The same restriction is found in the formation of the displacement

construction and wh-questions, both of which involve a displaced nom-
inal element that appears in the clause-initial position. There are two
types of such structure. The first type, termed the nominal type, is
constructed as an equational sentence in which the displaced RP or the
wh-word and the remaining elements of the clause are juxtaposed. This
remaining clause is preceded by a nominative case marker, and it is
structured like a headless RC. The second type, termed the verbal type,
is formed simply by placing an RP or a wh-word at the beginning of the
clause. The remaining clause of the verbal type stays structurally
unchanged. The wh-word can even appear in situ in the verbal type

though it more often appears clause-initially. It is the nominal type that

involves a restricted neutralization of semantic roles; that is, its pivot has
to be an actor of a gapped AV clause or an undergoer of a gapped UV
clause. As for the verbal type, the restricted neutralization is not found.
The nominal type is illustrated in (20) while (21), repeated from (10a′),
exemplifies the verbal type.17 The displaced RP is underlined and its pivot
is indicated by ‘___’ in the clause.
(20) Amis sentences with displaced RPs (the nominal type)
a. Pivot: (Patient) undergoer of an applied UV verb

O fafahian a singsii k-o ka-olah-an¼ako ___i.
cn woman lnk teacher nom-cn nfin-like-la¼1s.gen
‘It is female teachers that I like better.’
b. Pivot: Actor of AV verb
Ya wawai k-o mi-pa-nanom-ay ____i t-o
that child nom-cn av-caus-water-fac dat-cn
kolong.
water.buffalo
‘It is that child who fed water to the water buffalos.’
b′. Pivot: Actor of UV verb
*Ya wawai k-o ma-pa-nanom-ay ____i k-o
that child nom-cn uv-caus-water-fac nom-cn
kolong.
water.buffalo
‘It is that child who fed water to the water buffalos.’
c. Pivot: (Patient) undergoer of an applied UV verb
Ya nanomi k-o mi-pa-nanom-an¼to¼ako ci
that water nom-cn la-caus-water-la¼asp¼1s.gen ppn
mama-an ____i.
father-dat
‘That water is what I gave father to drink.’
(21) Amis sentences with displaced RPs (the verbal type)
Pivot: Oblique core argument of three-place UV verb

Ya nanomi ma-pa-nanom¼to n-u wawa k-o
that water uv-caus-water¼asp gen-cn child nom-cn
kolong ____i.
water.buffalo
‘The child has already fed the water buffalo that water.’
In (20a), the undergoer RP appears in sentence-initial position and there

is a gap in the remaining clause that follows the displaced RP. As one can
see, there is a case marker ko between the displaced RP and the
remaining clause; that is, the clause appears in a nominal position.
Furthermore, the verb inside the nominal clause is an applied UV verb.
Examining the rest of the examples in (20), we can see that they

968 JOY J. WU
demonstrate a restricted neutralization of semantic roles, as the pivot in

the nominal clause following the displaced element has to be the actor of
an AV verb or the undergoer of a UV verb; the latter can be either a plain
UV verb or an applied UV verb.
The restricted neutralization exemplified in (20) is also observed in the
nominal type wh-questions in (22):
(22) Amis wh-questions (the nominal type)
a. Pivot: Actor of AV verb

Cimai k-o mi-palo-ay ____i t-o wawa?
who.nom nom-cn av-beat-fac dat-cn child
‘Who is the one that beat the child?’
a′. Pivot: Actor of UV verb
*Cimai k-o ma-palo-ay ____i k-o wawa?
who.nom nom-cn uv-beat-fac nom-cn child
‘Who is the one that beat the child?
b. Pivot: Undergoer of UV verb
O maani k-o ma-kaen-ay n-i Aki ____i?
cn what nom-cn uv-eat-fac gen-ppn Aki
‘What is it that Aki ate?’
b′. Pivot: NMR direct core argument of AV verb
*O maani k-o k<om>aen-ay Ø-ci Aki ____i?
cn what nom-cn <av>eat-fac nom-ppn Aki
‘What did Aki eat?’
c. Pivot: (Patient) undergoer of applied UV verb
Cimai k-o ka-olah-an¼iso ____i?
who.nom nom-cn nfin-like-la¼2s.gen
‘Who is the one you like?’
The sentences in (22a–a′) exemplify wh-questions concerning an actor of a

predicate. As shown in the data, the clause following the interrogative pro-
noun is preceded by a case marker, which gives the nominal property of the
clause. Furthermore, when the interrogative pronoun functions as the actor of
the predicate, the verb has to be marked by the AV affix; this pronoun cannot
be interpreted as functioning as the actor of a UV verb. When the interrogative
pronoun refers to a non-actor in the clause, the verb has to be marked by either
the plain UV markers (e.g. (22b)) or the applicative markers (e.g. (22c)). Hence,
there is a restricted neutralization of semantic roles.18
26.5.2 Voice Operations and Applicative Constructions

Table 26.1 summarizes the voice system in Amis. As shown, Amis distin-
guishes three voices: neutral, actor and undergoer. Neutral voice is used for
marking verbs that are both M-intransitive and S-intransitive as their single
macrorole is not specified. The neutral voice will be further explored in
Section 26.6. The other two voices will be introduced in more detail here.

Table 26.1 Voice markers and the applicative markers in Amis (adapted from Wu
2006b: 289)
Voice Macrorole of the PSA Affixes

Neutral voice19 Unspecified mi-, ma-, <om>
Actor voice Actor mi-, ma-, <om>
Undergoer voice Plain Undergoer (unmarked choice) ma-, ma-. . .<om>,
ma-ka-, -en
Applicative Undergoer (marked choice) sa-, -an
Let us begin with the discussion of actor voice. AV verbs always have a
nominative-dative case pattern, and they have only one macrorole (i.e.
actor). The actor is assigned the nominative case and the NMR argument is
marked by the dative case. The voice-marking function of the AV affixes is
demonstrated in their co-occurrence with the volitative mood marker -aw,
which shows the UV pattern when it attaches to a root form.
(23) -aw volitative construction
a. Nanom-aw¼ho¼ako.
water-vol¼asp¼1s.gen
‘I will go drink water first.’ (Volitative mood, UV)
a′. Mi-nanom-aw¼ho kako.
av-water-vol¼asp 1s.nom
‘I will go drink water first.’ (The water is farther away than the one
mentioned in (23a).)
b. Kaen-aw¼ako k-o dateng.
eat-vol¼1s.gen nom-cn vegetable
‘I will try that vegetable.’ (Volitative mood, UV)
b′. K<om>aen-aw k-o wawa t-o sapaiyo.
<av>eat-vol nom-cn child dat-cn medicine
‘(I am) afraid that the child will take the medicine.’
c. Olah-aw¼ako kiso?
like-vol¼1s.gen 2s.nom
‘May I go to love you?’ (Volitative mood, UV)
c′. Ma-olah-aw kako tisonan.
av-like-vol 1s.nom 2s.dat
‘I am afraid that I will like you.’
As shown in (23), the suffix -aw manifests an optative reading for the derived
verb. Notice that the case-marking pattern for V-aw follows the UV pattern,
as the actor is marked by the genitive case. However, when the V-aw forms
are affixed with mi-, <om>, and ma-, their case-marking patterns become the
AV pattern, as indicated in (23a′), (23b′) and (23c′). This contrast shows the
voice-marking function of the AV markers. But what kinds of functions do
the AV constructions perform?
Clearly, the AV construction has a PSA-modulation function, as it makes a
marked choice of PSA in terms of the PSA Selection Hierarchy. Given the fact

970 JOY J. WU
that Amis displays ergative features in at least the case-marking system and
some constructions that involve a PSA such as the relative clause and the
nominal type of wh-question, one would expect the lowest-ranking argument
to be the unmarked PSA choice. However, in the AV construction, it is the
highest-ranking direct core argument that is chosen to be the PSA.
What about the argument modulation function? For a two-place predi-
cate, the lowest-ranking direct core argument in the AV sentences should be
assigned an undergoer based on the macrorole assignment principles, as
such verbs can take at most two macroroles. However, this argument in the
AV construction does not surface as a macrorole syntactically, as revealed by
its case marking and the fact that its status can be promoted by the appli-
cative construction. Instead, this argument is realized as an NMR core
argument in the AV construction. In other words, the lowest-ranking argu-
ment of a two-place predicate has been stripped of its macrorole status by
the AV operation. However, it is still in the core, as indicated by its behav-
ioural property in serving as a semantic controller in the persuade-type
control construction, as discussed in Wu (2006a). Hence, AV constructions
also perform an argument-modulation function. This function is even more
salient for three-place predicates, as a possible undergoer can also be
marked by the preposition in the AV construction in addition to the dative
case. Consider the following examples:
(24) Amis three-place predicates
a. Pa-caliw Ø-ci Kacaw t-o singsi t-o

caus-borrow nom-ppn Kacaw dat-cn teacher dat-cn
payso.
money
‘Kacaw lent the teacher money.’ (Causative, AV)
b. Pa-caliw Ø-ci Kacaw t-o payso i
caus-borrow nom-ppn Kacaw dat-cn money prep
singsi.
teacher
‘Kacaw lent the money to the teacher.’ (Causative, AV)
c. Aka pa-caliw-en k-o singsi t-o payso!
neg.imp caus-borrow-uv nom-cn teacher dat-cn money
‘Don’t lend the teacher money!’
As shown in (24), the recipient RP singsi can be marked either by the dative
case or the preposition in the AV construction. This RP is the second-highest-
ranking argument in the LS of pa-caliw ‘lend’, and it is also a possible under-
goer, as indicated in the UV sentence in (24c). The presumed undergoer RP is
realized as non-macrorole in the AV construction in (24a), but it is realized as
an adjunct in (24c), as the preposition i typically marks a locative RP in the
periphery. From the above discussion, we can thus conclude that the actor
voice not only modulates the PSA choice but also modulates the semantic
status of a core argument by either stripping a macrorole argument of its
macrorolehood or realizing a core argument as an oblique element.

What about the undergoer voice? As shown in Table 26.1, there are two sets
of UV markers: the marked set and the unmarked set. The markers that
signify a marked undergoer choice used to be treated as voice markers as
well (termed instrumental voice and locative voice) in Amis, but the examples
in (25) show that they serve functions other than that of a voice operation.
(25) The applicative UV constructions in Amis
a. Aka sa-pi-litek-en k-o-ra caklis

neg.imp ia-nfin-chop.tree-uv nom-cn-that axe
t-o-ra kilang!
dat-cn-that tree
‘Don’t use that axe to chop down the tree!’
b. Ma-sa-pi-sanga n-i Aki t-o takid k-o-ya
uv-ia-nfin-make gen-ppn Aki dat-cn bottle nom-cn-that
aol.
bamboo
‘Aki used that bamboo to make the bottle.’
The sentences in (25) show that when the UV marker -en or ma- and the
instrumental applicative marker sa- co-occur in a sentence, only the
instrument RP surfaces as the undergoer instead of the patient RP, which
would be the default undergoer following the AUH. In other words, the
applicative markers indicate a marked undergoer selection. The voice
markers and the applicative markers show different operations in the two
phases in the RRG linking algorithm. That is to say, the applicative marker
affects the linking from argument positions to macroroles, while the voice
marker operates at the linking from macroroles to syntactic functions.
There are two functions of these applicative markers. First, they may
enhance the status of a non-argument such as instrument or location to
become a core argument. Second, they can also promote a non-macrorole
core argument (e.g. patient in an AV sentence) to become a macrorole. The
instrumental applicative construction serves the first function, while the
locative applicative construction can perform both functions.20
Based on the discussion so far, the UV pattern should be deemed the
default pattern in Amis, which is proven by the fact that it is the unmarked
voice of the applicative constructions even when the UV markers do not
show up. Although the UV pattern enjoys unmarked status in Amis, there
are some predicates that seem to take the AV pattern by default, and for
such predicates, UV appears to be the marked pattern. Such predicates can
be illustrated by the pa- verbs. Consider the following examples of a pa- verb
plus the volitative mood suffix -aw:
(26) pa- verbs suffixed with the volitative marker -aw
a. Pa-nanom kako t-o kolong.

caus-water 1s.nom dat-cn water.buffalo
‘I feed water buffalos water.’ (Causative, AV)

972 JOY J. WU
b. Pa-nanom-aw¼ho¼ako k-o kolong.

caus-water-vol¼asp¼1s.gen nom-cn water.buffalo
‘I will feed the water buffalo water first.’ (Volitative, UV)
c. Mi-pa-nanom-aw¼ho kako t-o kolong
av-caus-water-vol¼asp 1s.nom dat-cn water.buffalo
‘I will go to feed water buffalos water first.’
Recall that in the previous discussion, I showed that when a root form is
suffixed with -aw, it takes the UV pattern. As we can see in (26a), the pa-
predicates appear with the AV case-marking pattern (i.e. nominative-dative).
However, when they are suffixed with -aw, the case-marking pattern
becomes the UV pattern. In other words, the pa- predicates behave like a
bare root form in the volitative mood construction. When the volitative
form pa-nanom-aw is prefixed with mi-, the case pattern becomes the AV
pattern again. These examples show that, unlike mi-, pa- does not have a
voice-marking function. However, pa- verbs follow the AV pattern by default.
To make pa- verbs appear in the UV pattern, the plain UV markers or the
applicative forms have to be used. Morphologically, the AV pattern appears
to be the default pattern of pa- verbs, while the UV pattern is a marked one.
However, syntactically, the UV forms actually turn a marked pattern (i.e. AV)
into an unmarked one. This may explain why the UV form pa- . . . -en and the
applicative form pa- . . . -an are found much more frequently than the plain
pa- forms in Amis.21
As mentioned, AV constructions perform both PSA-modulation and
argument-modulation functions. How about UV constructions? Though
appearing to be the basic pattern of Amis based on the case marking and
the default voice choice of the applicative constructions, they turn out to
be the marked voice choice for some predicates that usually appear with
the AV pattern by default. For such predicates, their plain UV constructions
perform a PSA-modulation function but no argument-modulation function
as the macrorole arguments and the NMR core arguments remain
unaffected. See Wu (2006a) for more discussion. However, the applicative
UV constructions display both functions, as now an adjunct becomes both
a macrorole and the PSA. The above discussion shows that both the actor
voice and the undergoer voice are deemed basic voice forms.22 Therefore,
Amis exhibits a split system in verbal morphology in spite of displaying
ergative features in the case-marking system and in some grammatical
constructions.
26.6 Special Discussion: An RRG Account for One-Place

Predicates in Amis
One-place predicates in Amis usually appear in an unaffixed manner (e.g.

tayra ‘go’ and miming ‘small’), or with affixes formally identical with the AV
markers such as <om> and ma- (e.g. t<om>angic ‘cry’ and ma-lalok

‘diligent’). In the RRG analysis proposed here, the voice marker that
appears on one-place predicates is treated as neutral voice as the only
macrorole of such verbs is not specified. This analysis is rather different
from the previous studies, which seem to assume a unified semantic role
for the single argument of intransitive predicates. For example, they either
label intransitive predicates as AV verbs (e.g. Liu 1999) or claim that the
single argument of intransitive verbs is patient (e.g. Chen 1987). In fact,
treating one-place predicates as AV or AF (i.e. agent-focus) verbs has been a
fairly common practice in the studies of other Formosan languages as well
(e.g. Atayal, as seen in Huang (2000) and Yeh (2015), and Tsou in Zeitoun
(1993), to name just a few). However, this unified-semantic-role approach is
implausible if we consider the rather different morphosyntactic behav-
iours of the single argument of the one-place predicates in Amis. In the
following, I am going to show how the RRG non-unified macrorole
approach for one-place predicates (Van Valin and LaPolla 1997; Van Valin
2005) can account for their peculiar morphosyntactic behaviours in two
constructions: the UV -en verbs and the evaluative construction.
26.6.1 The -en Form of One-Place Predicates

The first piece of evidence that proves the different macrorole assignments
of the sole argument of the one-place predicates comes from their corres-
ponding UV -en form, in which the actor gets the genitive case while the
undergoer takes the nominative case. Consider the following examples:
(27) Amis verbs with one core argument
a. T<om>ireng cingra.
<neut>stand 3s.nom
‘He is standing.’
a′. Tireng-en¼ako pa-kimad, ta paka-nengneng
stand-uv¼1s.gen caus-speech so.that ablt-see
kamo.
2p.nom
‘I will stand up when making a speech so that you can see (me) clearly.’
b. Ma-toni’ k-o-ni a titi.
neut-soft nom-cn-this lnk meat
‘This meat is soft.’
b′. Toni’-en¼ako k-o-ni a titi.
soft-uv¼1s.gen nom-cn-this lnk meat
‘I will soften this meat.’
Both t<om>ireng ‘stand’ and ma-toni’ ‘soft’ are traditionally labelled as AV

verbs. When they are suffixed with -en, the only argument in t<om>ireng
(now tireng-en) is marked by the genitive case. However, the single argument
in ma-toni’ (e.g. koni a titi) is still marked by the nominative case in the UV
construction toni’-en. The factor affecting the case-marking pattern is the

974 JOY J. WU
different macroroles assigned to the only arguments of t<om>ireng ‘stand’

and ma-toni’ ‘soft’. As the LS of t<om>ireng is do′ (x, [stand′ (x)], the x-
argument will be an actor. But, as there is no do′ in the LS of ma-toni’ (i.e.
(BECOME/INGR) soft′ (x)), the x-argument is an undergoer. When the verb is
affixed by -en, the agentive UV marker, the actor in tireng-en is marked by the
genitive case by default,23 while the undergoer in toni’-en receives the nom-
inative case in this UV -en construction. The above examples indicate the
inadequacy of labelling both of the two verbs as AV or AF verbs. Instead, the
RRG approach can account for the distinctive nature of the semantic roles
played by the single arguments of the two verbs.
However, one may run into a problem upon the application of the rules in
(14), repeated as (28) below, for intransitive verbs suffixed with the UV
marker -en.
(28) Case assignment rules in Amis
a. Assign nominative case to the lowest macrorole argument in terms of the

PSA Selection Hierarchy.
b. Assign genitive case to the other macrorole argument.
As shown in tiring-en in (27a′), the single argument of an -en intransitive verb

is always marked by the genitive case. Applying rule (28a) to an -en intransi-
tive verb will yield the wrong case assignment. Therefore, another set of case
assignment rules for verbs marked by -en has to be postulated. These rules
are stated in (29):
(29) Case assignment rules for verbs marked by -en
a. Assign genitive case to the highest-ranking macrorole in terms of the PSA

Selection Hierarchy.
b. Assign nominative case to the other macrorole argument.
For two-place or three-place -en verbs, all three of the rules in (29) are
applicable. But for the one-place -en verbs, only (29a) and (29c) will apply,
as there is only one macrorole in such verbs. The examples in (30) illustrate
how the rules in (28) and (29) work in Amis.
(30) The application of the revised case assignment rules in Amis
a. Ma-olah kako ci Panay-an.

av-like 1s.nom ppn Panay-dat
‘I like Panay,’
a′. like′ (kako, Panay)
b. Ma-ka-olah¼ako Ø-ci Panay.
uv-nfin-like¼1s.gen nom-ppn Panay
‘I love Panay (secretly).’ Or, ‘Panay was loved by me.’

b′. like′ (ako, Panay). . ..BECOME like′ (ako, Panay)

c. Ma-stol kako t-o fekeroh.
neut-stumble 1s.nom dat-cn rock
‘I stumbled over on the rock.’
c′. stumble′ (kako)
d. Ma-stol n-o fekeroh kako.
uv-stumble gen-cn rock 1s.nom
‘The rock rolled to me and made me stumble.’
d′. [do′ (fekeroh, Ø)] CAUSE [BECOME stumble′ (kako)]
e. Ma-rohem¼to k-o-ra pawli.
neut-ripe¼asp nom-cn-that banana
‘That banana has become ripe.’
e′. (INGR/BECOME) ripe′ (pawli)
f. Rakat-en¼ako.
walk-uv¼1s.gen
‘I will walk (to do something.)’
f ′. DO (ako, [walk′ (ako)])
g. Rakat-en¼ako k-o-ni a kayakay.
walk- uv¼1s.gen nom-cn-this lnk bridge
‘I will walk past this bridge.’
g′. DO (ako, [walk′ (ako, kayakay)]) & [BECOME walked′ (kayakay)]
h. Pa-si-fana’ k-o singsi t-o wawa t-o
caus-have-knowledge nom-cn teacher dat-cn child dat-cn
n-o ’Amis.
gen-cn Amis
‘The teacher is going to teach the children Amis.’ (Causative, AV)
h′. [do′ (singsi, Ø)] CAUSE [BECOME have.knowledge′ (wawa, no ’Amis)]
26.6.2 The Evaluative Constructions

RRG’s non-unified treatment of the semantic role played by the single
argument of the one-place predicates can not only account for the different
case-marking patterns of the one-place -en verbs discussed in the previous
section but can also offer a better explanation for the evaluative
construction, which can render two types of English sentences such as
‘The food tastes good’ or ‘It is difficult to do this job’. Amis examples for
the two types are given in (31) and (32) respectively.
(31) Amis evaluative constructions: Type I
a. Tada-fangcal a nengneng-en k-o-ra coka.

very-good lnk see-uv nom-cn-that picture
‘That picture looks really beautiful.’

976 JOY J. WU
b. ’angerer-ay a kaen-en k-o karokot.

bitter-fac lnk eat-uv nom-cn bitter.gourd
‘Bitter gourds taste bitter.’
(32) Amis evaluative constructions: Type II
a. Koesit a sowal-en ciira.

difficult lnk say-uv 3s.nom
‘It is difficult to talk to him.’
b. Tati’ih a fo’enot-en k-o fanoh n-o howak.
bad lnk pull.out-uv nom-cn feather gen-cn duck
‘It is difficult to pull out the feathers of ducks.’
c. Sa-koesit sa a tedal-en k-o-ni saso’ot.
so-difficult like.this lnk untie-uv nom-cn-this knot
‘It is so difficult to untie this knot.’
The Amis sentences in (31) and (32) all begin with a state predicate (i.e. the
evaluative predicate), which is then followed by the linker a and a UV verb. The
difference between the two types of evaluative construction is that the nom-
inative argument is a core argument shared by both predicates in (31) while the
one in (32) is an argument of the second verb but not the first one. Here we will
limit our discussion to the first type. A similar construction has been found in
other Formosan languages such as Seediq (Tsukida 2005) and Atayal (Yeh 2015).
Atayal examples are given in (33) (Yeh 2015: 132, original gloss).
(33) Atayal evaluative constructions24
blaq niq-un qu’ mami’ qa’.

good[av] eat-uv nom rice dem
‘This (type) of rice tastes good.’
Yeh analyses the first predicate of such constructions as an AV verb in spite

of the fact that it is a state predicate, and she argues that the construction is
a fusion of a serial verb construction (SVC) and a commentative complement
clause construction (commentative CCC) (Yeh 2015: 132). Her analysis is
summarized in Table 26.2 (cited from Yeh 2015: 148).
Table 26.2 The relationship between the juxtaposed verbs in the BLAQ evaluative
construction, SVCs and commentative CCCs in Squlip Atayal: a comparison (adapted
from Yeh 2015: 148)
Construction type BLAQ evaluative Commentative

parameter SVCs construction CCCs
(A) Manifestation of voice in verb (a) AV UV AV UV (a’) AV AV

sequences (b) UV AV (b’) AV UV
(B) Sharing of arguments Obligatory Obligatory Optional
(C) Sharing of TAM (tense-aspect- Obligatory Obligatory Depends
modality) information
(D) Sharing of polarity value Obligatory Obligatory Optional

According to the table, the Atayal evaluative construction shares more

features with SVC, but it is argued to also show the feature of the comple-
ment clause in its AV–UV sequence. Amis examples analogous to the Atayal
SVCs discussed in Yeh (2015) are provided in (34). Similar to those Atayal
examples, the second verb in such sentences has to be an AV form or an
intransitive verb with an actor (e.g. (34a), but there is no restriction on the
voice of the first verb.
(34) Amis SVC examples
a. Ma-herek¼to kako a ma-lafi.

av-finish¼asp 1s.nom lnk neut-dinner
‘I finished eating dinner.’
b. Ma-rara k-o fafahian a mi-fihon.
neut-slow nom-cn woman lnk av-put.on.makeup
‘The woman puts on make-up slowly.’
c. Tanam-en¼ako a pa-rakat k-o-ra tosiya.
try-uv¼1s.gen lnk caus-walk nom-cn-that car.
‘I will try to drive that car.’
Wu (2006a) mentions that these constructions all have an actor pivot,

either actor of an intransitive verb or that of an AV verb, in the
linked core.
Like the Atayal examples discussed in Table 26.2, the evaluative construction
in Amis also shares the following two features with SVCs: sharing of TAM
information and sharing of polarity value. First, in the evaluative construction
as well as in the SVC, only the first verb can be marked with TAM information.
(35) Sharing of TAM information in SVCs and the evaluative constructions in Amis
a. Mi-sawad¼to kako a mi-mali

av-quit¼asp 1s.nom lnk av-ball
‘I have quit playing ball.’
b. *mi-sawad¼to kako a mi-mali¼to
av-quit¼asp 1s.nom lnk av-ball¼ asp
c. Fangcal¼to a nengneng-en k-o-ra coka.
good¼asp lnk see-uv nom-cn-that picture
‘That picture looks good now.’
d. *Fangcal¼to a nengneng-en¼to k-o-ra coka.
good¼asp lnk see-uv¼asp nom-cn-that picture
Second, only the first verb in the named constructions can be conjugated
with the non-finite markers when following the negator caay ‘not’.25
(36) The negative sentences of SVCs and the evaluative constructions in Amis
a. Caay ka-safon-en nira k-o tireng a mi-ngingoy

neg nfin-soap-uv 3s.gen nom-cn body lnk av-shower
‘He didn’t use soap to wash (his) body.’

978 JOY J. WU
b. *Caay ka-safon-en nira k-o tireng a pi-ngingoy

neg nfin-soap-uv 3s.gen nom-cn body lnk nfin-shower
c. Caay ka-fangcal a nengneng-en kora coka.
neg nfin-good lnk see-uv nom-cn-that picture
‘That picture does not look good.’
d. *Caay ka-fangcal a ka-nengneng-en kora coka.
neg nfin-good lnk nfin-see-uv nom-cn-that picture
However, unlike the SVCs in (34), it is the second verb in the evaluative
construction that controls the case-marking pattern of the sentence, but
not the first verb. As shown in (34c), even though the second verb appears
in the AV form, the arguments in the linked core are case-marked with
the UV pattern, following the first verb. In other words, the second verb
in the SVCs has no voice function. Nevertheless, the arguments of the
linked core in the Amis evaluative constructions always follow the case-
marking pattern of the UV verb, as seen in the comparison between (37a)
and (37b).
(37) The case-marking pattern of the Amis evaluative construction
a. Tada-fangcal a nengneng-en¼ako k-o-ra coka.

very-good lnk see-uv¼1s.gen nom-cn-that picture
‘I saw the picture and thought that it was very beautiful.’
b. *Tada-fangcal kako a nengneng-en k-o-ra coka.
very-good 1s.nom lnk see-uv nom-cn-that picture
c. Tada-fangcal k-o-ra coka a nengneng-en.
very-good nom-cn-that picture lnk see-uv
‘That picture looks really beautiful.’
Yeh (2015) in particular explores the answer to the question of why the
second verb in the evaluative construction has to be marked by the UV form,
but not the AV. She proposes that it is the undergoer that is evaluated in the
construction, and this pragmatic function accounts for the UV marking of
the second verb. However, the RRG approach might offer an even more
straightforward answer.
Following the RRG analysis, the first verb in (31) will have an undergoer
instead of an actor, and the voice sequence in the Amis evaluative con-
struction should be a neutral voice followed by a UV verb. The shared
arguments in such sentences play the role of undergoer (undergoer of
the intransitive verb and undergoer of a UV verb). Hence, it is quite
reasonable to maintain the syntactic status of the undergoer in both
cores. If the second verb takes the actor voice, the macrorole status of
the undergoer will be removed, as we have seen in the function of the
voice discussed earlier. The seeming paradox in Yeh’s study lies in her
treatment of the only argument of the first verb as an actor or a unified
thematic role, but the RRG perspective can provide an account that
resolves the paradox.

As for the SVC examples, the shared argument might play different
semantic roles in the two linked verbs, but this role is always an actor in
the second verb. This may explain why the second verb bears AV marking
though the AV does not have any voice function.
References
Blust, Robert. 1999. Subgrouping, circularity and extinction: Some issues

in Austronesian comparative linguistics. In Elizabeth Zeitoun and Pau
Jen-kuei Li (eds.), Selected Papers from the Eighth International Conference on
Austronesian Linguistics, 31–94. Taipei: Institute of Linguistics, Academia
Sinica.
Bril, Isabelle. 2016. Information structure in Northern Amis:
A morphosyntactic analysis. Oceanic Linguistics 55(2): 449–479.
Chen, Teresa. 1987. Verbal Constructions and Verbal Classification in Nataoran-Amis
(Pacific Linguistics C-85). Canberra: The Australian National University.
Guerrero Valenzuela, Lilián and Robert D. Van Valin, Jr. 2004. Yaqui and the
analysis of primary object languages. International Journal of American Lin-
guistics 70: 290–319.
Huang, Lillian. 2000. Verb classification in Mayrinax Atayal. Oceanic Linguis-
tics 39(2): 364–390.
Jolly, Julia. 1993. Preposition assignment in English. In Van Valin (ed.),
275–310.
Liu, Dorinda Tsai-hsiu. 1999. Cleft Constructions in Amis. MA thesis, National
Taiwan University, Taipei.
Liu, Emma En-hsin. 2003. Conjunction and Modification in Amis. MA thesis,
National Tsing Hua University, Hsinchu.
Starosta, Stanley. 1974. Causative verbs in Formosan languages. Oceanic
Tsuchida, Shigeru. 1988. Amis. In Takashi Kamei, Rokuro Kono and Eiichi
Chino (eds.), The Sanseido Encyclopedia of Linguistics, Vol. 1: Languages of the
World, Part One, 447–449. Tokyo: Sanseido Press.
Tsukida, Naomi. 2005. Seediq. In Alexander Adelaar and Nikolaus P.
Himmelmann (eds.), The Austronesian Languages of Asia and Madagascar
(Routledge Language Family Series), 291–325. London: Routledge.
Wang, Samuel Hsu. 1976. The Syllable Structure of Fataan-Amis. MA thesis,
National Taiwan Normal University, Taipei.

980 JOY J. WU
Wu, Joy Jing-lan. 2003. Clausal modifiers in Amis. Concentric: Studies in English
Literature and Linguistics 29(2): 59–81.
Wu, Joy Jing-lan. 2006a. Verb Classification, Case Marking, and Grammatical
Relations in Amis. PhD dissertation, University at Buffalo (SUNY).
Wu, Joy Jing-lan. 2006b. The analysis of pa- verbs in Amis. In Henry Y.
Chang, Lillian M. Huang and Dah-an Ho (eds.), Streams Converging into an
Ocean: Festschrift in Honor of Professor Paul Jen-kuei Li on his 70th Birthday
(Language and Linguistics Monograph Series Number W-5). Taipei: Aca-
demia Sinica.
Yeh, Maya Yu-ting. 2015. An atypical SVC? A study of BLAQ UV qu’ NP
construction in Atayal. In Elizabeth Zeitoun, Stacy F. Teng and Joy J. Wu
(eds.), New Advances in Formosan Linguistics, 131–155. Canberra: Asia-Pacific
Linguistics.
Zeitoun, Elizabeth. 1993. A semantic study of Tsou case markers. Bulletin of
the Institute of History and Philology 64(4): 969–989.
Notes
1 www.apc.gov.tw/portal/docList.html?CID¼7AD36169AB07E1D0.
2 The phonetic symbols used in the transcription generally follow the IPA
system, with the following exceptions: /e/ stands for schwa [ ], /d/ for e
voiceless lateral [ɬ], /’/ for epiglottis stop /ʡ/, and /ng/ for /ŋ/. The glosses in
general follow Wu (2006a), which adopts some of Liu’s (1999) morphemic
analyses (e.g. the separation of noun classifiers and case markers) with
some slight modification.
3 There are two classes of nouns identified in Amis: personal proper nouns,
including people’s names and kinship terms, and common nouns. The
noun class marker usually appears with the case marker and forms a
case-marking composite, but the case marker does not show up in
predicate position.
4 The analysis of the prefixes ka- and pi- as markers of non-finite forms
follows Bril (2016), but Bril specifies pi- as a non-finite marker for actor
voice verbs (i.e. NFIN.AV). Here, pi- is simply glossed as NFIN.
5 However, Wu employs an ideophone-forming construction and classifies
the root forms into five classes: object, state (attribute), state (transient/
result), activity, achievement/semelfactive (Wu 2006a: 156).
6 As argued in Wu (2006a), the voice markers carry both inflectional and
derivational function, and the latter function changes not only the
lexical categories of the attached roots or stems but also their semantics.
These voice markers also have their own decomposed structures. For
example, a UV verb does not merely enhance the status of a patient role
as it would in an analogous English passive sentence. It also has a

different meaning due to the UV marker ma- and the derived verb might
become an (active) accomplishment verb.
7 The factual marker -ay also carries a nominalizing function, which
roughly means ‘someone who . . .’ or ‘something which . . .’.
8 Exceptions are found in state predicates with a gradable feature (e.g. ma-
lalok ‘diligent’); they can readily appear after a case marker in a compara-
tive construction. See Wu (2003, 2006a) for details.
9 Examples (8b) and (8c) exemplify a nominal type of wh-question . Only
the PSA can be the missing argument in the nominal clause. This will be
further discussed in Section 26.5 on grammatical relations.
10 For a special type of three-place predicate that is prefixed by pa-pi-,
Principle B is the only one that is applied.
11 To simplify the discussion, the LS of ma- (active accomplishment, UV) is
not represented in the LS of the ma- UV construction of the three-
place predicates.
12 The semantic role of the noun marked by the dative case is indicated in
parentheses after the translation.
13 The differences between the plain UV and the applicative UV construc-
tions are subtle. In general, the -en plain UV construction emphasizes the
volition of the actor, and it is more likely to refer to a future event; it is
also the form that is used in imperative sentences. The ma- plain UV
construction highlights the affectedness of the undergoer. The mi- . . . -an
applicative UV stresses the identity of the actor.
14 The to RP in (12a) in fact manifests an external causer for the event
described by the predicate, though the predicate is non-causative. The
same phenomenon is also found in the to RP in (12c). This explains why
they can serve as the argument for the mi- counterparts in (13), which
carry a causative reading after derivation. Functionally speaking, the to
marker here is similar to the English preposition from, which appears to
be causative in its predicative roles (Jolly 1993: 293), as in the sentence
John died from Malaria. The causative version of this English sentence
would be Malaria killed John.
15 This set of rules cannot work for the verbs suffixed by the UV marker -en.
We will therefore propose another set of rules for such verbs in the
special discussion of Amis one-place predicates in a later section of
the chapter.
16 The RC is boldfaced in the examples. The ‘__’ and the coreferencing
subscript in the examples is for expository purposes only. It does not
refer to a trace or a null copy in the site of the missing RP. RRG does not
posit such notions.
17 Choosing one type over the other crucially depends on the status of
the displaced RP. If it is a core argument, then the nominal type is
preferred or even required for some speakers; if it is an oblique
argument or adjunct, then the verbal type is allowed. As only the

982 JOY J. WU
nominal type involves a grammatical relation here, we only give

details of this type.
18 Notice that the verbal type is also found in the formation of wh-ques-
tions and the criterion for choosing the nominal or the verbal one in
forming a wh-question is roughly the same as that for a displacement
construction.
19 This voice marker is used for intransitive verbs of which the only macro-
role is not specified. Such verbs used to be termed actor voice (or agent
voice/focus) in most of the previous studies of Amis and other
Formosan languages.
20 An interesting feature of these applicative forms is that they can all be
used as nouns designating the argument that is affected by them, and
some may even be lexicalized. For example, the sa- applicative form can
usually refer to an instrument or a reason, while -an applicative form can
designate an object that is acted upon (e.g. mi-tilid-an ‘something written’
> mi-tilid ‘write; study’) or a location (e.g. pi-tilid-an ‘school’).
21 This has been pointed out by Starosta (1974) and my investigation con-
firms this finding.
22 Based on this proposal, I have maintained the terminology of actor voice
and undergoer voice in the discussion, instead of using undergoer voice
and antipassive voice, or actor voice and passive voice.
23 The nominative argument of the first part of (27a′) is not expressed. It
can either be the theme argument (e.g. koni, a nominative demonstrative
pronoun meaning ‘this (talk)’) or the recipient argument (e.g. kamo, a
2nd-person plural nominative pronoun meaning ‘you (all)’) of the second
core pakimad ‘give a talk’. Judging from the second part of (27a′), kamo
seems to be a more natural choice. The omission of this argument in the
first part could be to avoid repetition. However, tirengen cannot take a
nominative argument by itself without the second core (i.e. *Tirengen ako
koni/kamo. is unacceptable.) One can only say ‘Tirengen ako.’, meaning
‘I want to stand.’ For example, when asked to take a seat, one can answer
with ‘Tirengen ako.’ The reader should also be aware that the acceptability
of (27a′) varies among dialects in Amis.
24 Yeh named this construction BLAQ UV qu’ NP in her work (Yeh 2015). To
avoid the terminological confusion between NP and RP in RRG, this
construction will simply be named BLAQ Evaluative Construction in
this chapter.
25 Amis verbs change their forms after the declarative negator caay ‘not’
and the imperative negator aka. See Wu (2006a) for detailed discussion.

Index
aboutness semantic representation of, 411

condition, 625–626, 628 adverbial clause. See juncture–nexus
relation, 466 combinations, clausal subordination
topic, 462–463, 465, 469 agent, 9, 22, 107, 243–244
absolutive agreement, 63, 66, 72, 116, 119, 153, 278, 282, 299,
case, 120, 274, 296, 313, 323, 327, 329, 332, 430, 327, 375, 377, 508–509, 583, 657, 674, 862,
479, 656 869, 871, 896, 898, 915–916
case system. See case marking Aktionsart, 94, 99, 220, 242, 891
accessibility, pragmatic, 459 accomplishment, 96, 221, 306
accusative achievement, 95, 221
case, 119, 142, 271, 323–324, 328, 331, 335, 430, acquisition of, 669
510, 654, 660, 682, 871 active accomplishment, 96, 306
case system, 332, See also case marking activity, 95, 221, 306
Acehnese, 280, 327, 583, 681, 684 causative types, 95, 99, 302
acquisition process, 96
language, 19, 227, 651, 666–667, 686 semelfactive, 98
Operating Principles, 666 state, 95, 220
activation, 459–460, 464, 502, 504, 594–595, 864 Alacatlazala Mixtec, 298
Activity Hierarchy, 259 Albanian, 343, 650
actor. See macroroles Algonquian languages, 832
Actor-Undergoer Hierarchy, 108, 244, 251, 295, alignment, 258, 273, 656, See also case marking,
321, 432, 503, 565, 599, 670 privileged syntactic argument (PSA),
adjective, 28, 30–31, 33, 47, 51, 101, 183, 186, 191, Selection Hierarchy
221, 232, 390, 415, 862, 866, 890 ergative-absolutive, 275
adjunct. See also adposition nominative-accusative, 275
adpositional phrase, 430 split, 871
ideophone, 422 Amele, 374, 536, 876, 915–950
non-phrasal, 44–47 Amharic, 659
phrasal, 42–44 Amis, 954–979
adposition anchoring, pragmatic, 458
assignment rules, 435–436 Angami Naga, 283
layered structure of, 49, 430 animacy, 141, 143–144, 299, 343, 433, 479,
non-predicative, 49–50, 100, 188, 428, 434 826–827, 830
postposition, 49, 193, 196, 868, 890–891, 915, 922 anticausative construction, 304
predicative, 49, 100, 187, 428, 434–435 antipassive. See voice
adjunct, 101, 434 aphasia, 718–719
argument-adjunct, 101, 430, 442 applicative, 142, 248, 272, 300–303, 837–838, 868,
preposition, 49, 51, 189, 434–435, 890–891, 955 870, 880, 916, 918, 932, 958, 961, 965, 968
semantic representation of, 435–436 appositive, 59
adverb, 31, 33, 186–187, 403–404, 890, See also Arabic, 26, 234, 785, 787, 792–793, 796
adjunct, non-phrasal Arapaho, 832
acquisition of, 678 Archi, 160
adverb ordering, 407 argument, 21
ideophonic, 412 core, 22–23
mimetic, 412 semantic, 22

984 INDEX
argument (cont.) complementation, 69, 77, 561, 771, See also

sharing, 73, 550, 558 subordination
syntactic, 22 complementizer, 592–593, 876, See also clause
argument structure, 58, 107, 292, 559, 582, 683, linkage marker
721, See also logical structures Completeness Constraint, 116, 125, 293, 300, 336,
alternations, 293–294 596–597
argument-adjunct. See adposition, predicative extended, 496
Armenian, 650, 658 complex sentences, 68, 72, 76, 525, 617, See also
aspect, 33, 202, 406, 533, 825, 894, 904, 927 clause linkage, juncture–nexus
acquisition of, 667 combinations
assertion, pragmatic, 457, 466, 493, 618, 629 concessive clauses, 570
Atayal, 973, 976 conditional clauses, 570
Avatime, 889–911 conjunction reduction, 500, 569, See also topic,
topic chain
Bahasa Indonesia, 108 constituent projection, 33, 70, 115, 372, 377
Bakusu, 309 Construction Grammar, 544
Bambara, 597 constructional schema, 10, 125, 130, 161, 478,
Bantu languages, 66, 308–309, 312, 414, 474, 500, 569, 583, 725, 805
508–509, 868 contrastiveness, 471
Barai, 41–42, 148, 279, 536 control construction, 64, 151, 156, 158, 536, 581,
Barbareño, 562 748, 879, 970
Basque, 122, 328 obligatory, 159, 582
Belhare, 566 controller. See privileged syntactic argument
Bella Coola, 375 copula, 39, 185, 221, 602, 798, 829, 863, 895, 917
Blackfoot, 832 be, 101
borrowing, 659 Cora, 546
branching, 20, 36, 47, 58 core, 21
direction theory, 651 core argument, 5, See also argument, core
Broca’s area, 695 direct, 81
Bulgarian, 50, 343, 650, 657 oblique, 81, 100
Burmese, 54 Creek, 870
Croatian, 61, 118, 122, 439, 506, 647, 653, 659,
case marking, 119, 271, 318, 374, 428, 433, 509, 661, 711–714, 721
630, 654, 711, 873–874, 955 cross-reference, 64, 72, 319, See also agreement
accusative system, 323, 327, 657 Cushitic languages, 659
assignment rules, 323, 328
ergative system, 323 Danish, 618, 621
hierarchy, 339 dative alternation, 294, 298, See also variable
nominative system, 108 undergoer
syncretism, 339–340 dative case, 108, 119, 274, 319, 323, 433, 439–440,
Catalan, 346 660, 955
Caucasian languages, 153, 160 dative shift. See dative alternation
causation, 7, 89, 245, 250, 302, See also Interclausal degree achievement, 97
Relations Hierarchy, Aktionsart dependence
causative construction, 302, 546 operator sharing, 71
Cavineña, 572 vs. subordination, 72
Cheyenne, 824
ˇ dependent-marking languages, 64, 308, 377,
Chichewa, 66–68, 308, 414, 518 382
Chinese, 231, 270, 279, 281, 572, 618 derivational morphology, 62, 371
Mandarin Chinese, 29, 77, 93, 205, 275, 280, Dhivehi, 26, 122
467, 481, 494, 496, 721 Didinga, 415
Choctaw, 279–280, 870 direct/inverse system, 825, 871–872
Chuj, 583 directionals, 202, 531, 757, 896
clause Discourse Representation Theory, 481, 489, 492
layered structure of, 21, 23, 27, 430 ditransitive, 298–301, 869–870, 919
non-universal features of, 24 Djaru, 340–341
relative, 594, See also relative clause double-marking languages, 68, 328, 343
universal features of, 24, 429 Dutch, 126, 659, 788
clause linkage, 68, 525, 559 Dyirbal, 27, 31–32, 108, 122, 135, 208, 257, 271,
asymmetrical, 79 273–274, 279, 310, 313, 322, 566, 583, 684,
juncture, 69–71, 527 787
nexus, 71–75, 527
symmetrical, 77 Emai, 404, 415
clause linkage marker, 828, 910, 942 Enga, 273, 280
cleft sentence, 591, 602 ergative
it-cleft, 74, 507, 607 case, 120, 129, 160, 323, 327, 329, 331, 430, 479,
Comanche, 581, 583 656–657, 869–872, 960
comment, 490 case system, 327, 332–333, See also case marking

Index 985
ergativity, 340, 871 Hakha Lai, 299, 308

syntactic, 313, See also ergative, case system Halkomelem, 343
Estonian, 340, 343–344, 348–349, 351, 355 Head-Driven Phrase Structure Grammar, 3, 738,
evaluative construction, 975 788
event quantification, 202, 406, 414, 417, 757, 825, head-marking languages, 63, 295, 308, 370, 376,
926 378–379, 434, 504, 825, 862, 865, 893, 915
evidentials, 37, 182, 202, 406, 531, 679 Hebrew, 791
exceptional case marking. See matrix-coding Biblical Hebrew, 785, 787, 791
constructions and raising Hindi, 340, 346
extraction restrictions, 616, 618, 620 Huallaga Quechua, 532
Japanese, 621 Hungarian, 135, 304
relativization, 621
topicalization, 621 Icelandic, 24, 68, 108, 118, 140–141, 154, 322,
wh-questions, 621 324–326, 332, 566, 654–655, 697, 716
wh-question linking template, 710 iconicity, 88, 430, 742
identifiability, 458–459
Farsi, 26 ideophone, 404, 890, See also adverb
Finnish, 343–344, 350–351, 667, 681 illocutionary force (IF), 33, 41, 202, 406, 531, 568,
focus, 457, 459, 463–464, 605, 610 652, 685, 925
argument, 470 assertion, 35
contrastive, 24, 604, 844 immediate common ground, 53, 490
fronting, 475 incorporation, 22, 295, 308, 834, 915
lexical alternations, 478 incremental theme, 97
marking, 25–26, 39, 465–466, 472, 480, infinitival complement, 152, 349, 560, 711, 773
603 inflectional morphology, 62, 371, 377
narrow, 469, 477, 593, 847, 864 information structure, 114–116, 456–457, 488
marked, 471 information unit (IU), 115, 461, 489
unmarked, 471 instrument, 107, 243, 250–251, 387, 432, 437–438,
predicate, 466, 478, 844, 864 504, 869, 909, 924, 931, 961, 971
sentence, 468, 846 instrumental prefix, 64, 309
focus domain, 622 instrumental case, 108, 329–330, 334–335, 442,
actual, 467, 489, 508, 934 660, 838
potential, 467, 472–473, 477, 489, 508, 604, 619, Interclausal Relations Hierarchy, 7, 87–88, 545,
933 564, 907–908
focus structure, 466, 472, 843, 902 semantic hierarchy, 83–87, 545, 563, 877
complex sentences, 619 syntactic hierarchy, 82–83, 543, 548
projection, 489, 491, 618 intonation, 26–27, 477, 492, 897, 939
subordinate, 472, 625 Inuktitut, 340
focus-sensitive elements, 604 inverse construction, 827, 852, See also direct/
Formosan languages, 954, 973, 976 inverse system
Frame Semantics, 219 Iraqw, 528
Frankish, 659 Irish, 650, 785
French, 89, 105, 130, 155, 247, 253–255, 298, Old Irish, 653
467–468, 480, 529, 537, 541, 604, 649, 659, island constraints, 616–617, 748, See also
662 extraction restrictions
Middle French, 649 Italian, 47, 89, 256, 346, 467, 474–476, 478, 480,
Functional Grammar, 430 512, 566, 618
FunGramKB, 220, 227–228, 804
Jakaltek, 89, 108, 117, 126, 257, 280, 322, 566
gender, 374–375 Japanese, 26–27, 29, 42, 108, 110, 130, 147, 202,
Generative Lexicon Theory, 102, 218, 230 331–332, 412–414, 417–419, 465, 467, 481,
Georgian, 306, 322, 566 494–496, 537, 541, 547, 566, 572, 621–626,
German, 24, 43, 48–49, 51, 61, 68, 108, 118–119, 671, 677, 768, 785, 920
126, 187, 189, 247–248, 253–255, 257, 326, Old Japanese, 548, 647
505, 566, 659, 716, 787 juncture, 69–71, 92, 559, See also clause linkage
Germanic languages, 375, 652 clausal, 70, 527
gerund, 74, 79, 560, 584 core, 69, 528
Government and Binding, 19 nuclear, 69, 530
grammatical relations, 4, 72, 117, 119, 269–270, sentential, 70
272, 827, 854, 965 juncture–nexus combinations, 76, 82, 561
non-universality of, 4 clausal coordination, 76, 910
grammaticalization, 653 clausal cosubordination, 77
Greek, 343, 648, 654, 658 clausal subordination, 910
Classical Greek, 661 adverbial, 80, 571
Homeric Greek, 650, 661 daughter, 79
Greenlandic Eskimo, 297 complex RPs, 92–94
Gum languages, 915 core coordination, 156–157, 529, 905, 908
Guugu Yimidhirr, 105 core cosubordination, 91, 155, 538, 908

986 INDEX
juncture–nexus combinations (cont.) default assignment principles, 113, 246

core subordination, 73, 79, 88, 533–534, 561 undergoer, 109–110, 244
daughter, 150 variable linking, 138, 145, 248, 252, 506, 660, 726
peripheral ad-core, 150, 543, 575 Maithili, 566
nuclear coordination, 530 Malagasy, 592, 613
nuclear cosubordination, 77, 88, 207, 538, 586, Maltese, 186, 205
908 Mandarin, 105
nuclear subordination, 534, 939 Maori, 527, 785
sentential coordination, 539, 561, 574 Marathi, 661
sentential subordination, 539, 561 matrix-coding constructions, 324, 768
mimetics, 412, See also adverb
Kabardian, 153–154, 340–341, 352 Minimalist Program, 19
Kaluli, 479, 509 modality, 33, 182, 200, 202, 406, 536, 550, 653,
Kashmiri, 346 825, 887, 903
Kewa, 37, 204 acquisition of, 687
Kharia, 187, 190–198 mood, 33, 904, See also illocutionary force
Khwe, 533 Mparntwe Arrernte, 647
Kikuyu, 66–67, 509 Muskogean languages, 870–871
Kinyarwanda, 312, 322, 566
Kokama, 563, 579 Nahuatl, 295
Korean, 41–42, 204, 509–510, 566, 618, 680 natural language processing, 788, 811
Natural Serialization Principle, 6, 652
Lakhota, 22–23, 27–28, 30–32, 34–37, 39, 56, 63, negation, 33, 37, 41, 55, 202, 406, 825, 896, 904
65, 68–69, 87, 147, 273, 378, 509, 583, 617, Nepali, 566
621, 767, 772, 865 neutralization
Latin, 31, 433, 648, 650, 655, 657, 660 restricted, 272, 274–275, 277, 279, 681
layered structure of the clause. See clause, layered unrestricted, 275
structure of nexus, 71–75, 92, 559, See also clause linkage,
layered structure of the word, 62, 372–373 juncture–nexus combinations
operators, 373 coordination, 72, 74–75, 559–560, 765, 773
left-detached position, 26 vs. conjunction, 72
lexical decomposition, 94–95, See also logical cosubordination, 71–72, 74–75, 536, 560, 753,
structures 765, 944, 946
lexical entry, 95, 230, 236, 371, 384–385, 428, 703, subordination, 71, 75, 77, 771, 828, 876
707, 715, 786, 788, 809–810 ad-subordination, 80, 83, 558, 570, 944, 948
Lexical Functional Grammar, 3, 20 daughter, 79, 83
lexical integrity hypothesis, 63 nominalization, 862, 876–882
lexical rules, 306, 786, 788, 792, 808 nominative
lexicon, 10, 95, 114, 117, 191, 200, 218, 220, 225, case, 119, 129, 271, 323, 326, 328, 331, 430, 657,
227, 229, 231, 314, 371, 383, 386, 435, 695, 682, 868, 871
703, 715, 717, 722, 745, 785 case system. See case marking
Linguistic Knowledge Builder, 788 Northern Sámi, 343, 347, 351, 354
linking, 777 Norwegian, 234
acquisition of, 683 noun complement, 622
lexical phase of, 8, 147 noun phrase, 5, 28, 208, 282, 284, 298, 609, 867,
morphosyntactic phase of, 8, 147 See also reference phrase
processing model, 699–700 NP, 28, See also reference phrase
linking in complex sentences, 150, 568, 576, 582 nucleus, 21–22
algorithm, 150 number, 374–375, 867
cleft constructions, 607
complex RPs, 596, 598 object, 108, 269
linking in simple sentences, 116, 841 direct, 110
role of information structure in, 488 double-object construction, 111
semantics-to-syntax, 116, 123, 495, 565, 722, indirect, 110
874 of predicative PP, 134
algorithm, 116 primary object language, 248, 299, 841, 869
syntax-to-semantics, 123, 514, 566, 722 primary object pattern, 142
algorithm, 123 oblique constituent, 61
logical structures, 95, 99, 200, 243–244, 306, 320, core. See core argument
384 peripheral. See adjunct
conceptual, 228, 230 Ojibwa, 832
Luwo, 414–415 Old Church Slavonic, 50
Old English, 50, 282, 372, 648, 654
Macedonian, 343, 653, 657 Omaha, 647
Macro-Event Property, 87, 91 operator, 33, 36, 57, 201, See also reference phrase
macroroles, 108–109, 242, 244, 271, See also Actor- clausal, 202
Undergoer Hierarchy core, 202
actor, 108, 244 nuclear, 202

Index 987
ordering, 36, 202, 204 acquisition of, 684

semantic representation of, 207 in-situ wh-word, 473, 966
sharing, 71 yes/no questions, 58, 618, 934
operator projection, 34, 38, 40, 55, 202 acquisition of, 684
Optimality Theory, 331 quirky case, 325
Paez, 206 raising, 73, 151, 156, See also matrix-coding

Palauan, 340–341 constructions
Papuan languages, 868, 876, 915 Rawang, 282
parser, 123, 497, 699, 776, 787, 790 reanalysis, 658
passive. See voice reason clauses, 572
Pastaza Quechua, 414 reciprocal
periphery, 22, 405, 421, 435, 447 prefix, 868
Persian, 785 verb form, 944
Old Persian, 340, 342, 353 reference phrase (RP)
Person Hierarchy, 841 core-level periphery, 93
Piraha,
~ 83 juncture–nexus combinations
Pisaflores Tepehua, 311 core coordination, 93
Pitjantjatjara-Yankunytjatjara, 373 core cosubordination, 93
pivot. See privileged syntactic argument (PSA) nuclear coordination, 93
Plains Cree, 832 nuclear cosubordination, 93
Polish, 619–621, 672–678, 682 RP coordination, 92
polysynthesis, 825, 861 RP cosubordination, 92
possession, 114, 224, 243, 437, 899, 965 RP subordination, 93, See also relative clause
alienable, 929, 932 layered structure of, 53
change of possession verb, 111 logical structure, 209
existential verb, 105 operators, 53, 208
inalienable, 929 classifiers, 54
possessive RP, 516, 930 referent tracking, 270, 272, 279–280, 835
semantic representation, 246 reflexivization, 89, 133
possessor raising, 106, 868 anticausative, 305, 682
post-core slot, 26, 465, 566, 623, 628–629, diachrony, 650
923–924, 956 reflexive binding, 149, 716
post-detached position, 26, 187, 465, 577, 623, 923 role hierarchy, 716
postposing construction, 628 Relational Grammar, 3, 19–20
postposition. See adposition relative clause, 69, 594, 882
pre-core slot, 24–25, 38, 44, 57, 465, 471, 474, 623, Accessibility Hierarchy, 591
749, 809, 844, 846, 848, 893, 924, 956 externally headed, 591, 595
pre-detached position, 26, 410, 463, 465, 467, 623, free, 600
893, 923 internally headed, 592, 597
predicate, 21 non-restrictive, 599
preposition, 49, See also adposition restrictive, 595
presupposition, pragmatic, 457, 462, 490, 493 relativization, 625
Principles and Parameters, 3 resultative construction, 70, 104, 753
privileged syntactic argument (PSA), 117–118, Rhetorical Structure Theory, 86
272, 874 Riau Indonesian, 275, 280
controller, 119, 272, 276, 278 right-detached position, 26
pivot, 119, 272, 275 Romance languages, 652, 811
Selection Hierarchy, 118, 275, 322 Romani, 658
selection principles, 275 Romanian, 342
process, 97 Russian, 29, 31, 105, 108, 185, 255, 257, 334, 336,
processing, language, 696–699 475–476, 505, 660, 675
projection grammar, 740
constituent projection, 740 Sama, 126–129, 322, 566
operator projection, 740 Sama-Bajaw, 303
speech act projection, 467 Sanskrit, 661
pronouns, 61, 186 Vedic Sanskrit, 658
proximate/obviative system, 826, 835, 848 Seediq, 976
punctual, 95, 98 Semitic languages, 652, 659
Punjabi, 346 serial verb constructions, 904, 907, 937
purpose clauses, 579 Sesotho, 474, 508
Setswana, 474, 508
Qiang, 37, 284 Sicilian, 476–477
qualia, 102, 222–224, 384 sign languages, 802
quantifiers, 55, 208, 297, 928 Irish Sign Language, 802
questions Siwu, 414
wh-questions, 25, 134, 139, 473, 511, 617, 622, Slavic languages, 433, 660
630, 708 Somali, 659

988 INDEX
Southern Tiwa, 297 frame-setting, 463, 726

Spanish, 66–67, 231, 255, 305, 340, 346, 371, 377, referential, 462–463, 469
382, 384, 386, 393, 433, 436–437, 439–446, topic chain, 570
448–450, 573, 575–576, 578, 605–606, 785, topicalization, 624–625, 627
805 Toqabaquita, 583
Argentinian Spanish, 460, 657 transitivity, 112–113, 298
split brain, 694–696, 715 acquisition of, 682
St’át’imcets, 376 M[acrorole] transitivity, 112
stage topic, 469 S[yntactic] transitivity, 112
Standard Theory, 19 Tree Adjoining Grammar, 743, 813
state, change of, 95 Tree Wrapping Grammar, 752, 813
status, 37, 182, 202 Tsez, 153
subjacency, 617 Tsou, 973
subject, 25, 269–270, 272, 275, 348, 382, 657, 681, Tsova Tush, 110
See also privileged syntactic argument Tukang Besi, 530
dummy, 121 Tupinamba, 297
within RP, 57 Turkish, 37, 122, 204, 231, 303–304, 311, 529, 531,
subordinate clause. See juncture–nexus 671–672, 677, 686, 758, 771
combinations, nexus, subordination Tzotzil, 25
adverbial clause, 558 Tzutujil, 280
complement clause, 77
relative clause, 594 Udihe, 535–536
subordination. See nexus, juncture–nexus undergoer. See macroroles
combinations UniArab, 792
nominalization, 876 Ute, 126
Swedish, 618, 621 Uto-Aztecan languages, 140, 295, 546, 581
switch-reference, 279, 549, 863, 941, 944, 949
syntactic inventory, 60, 715, 738, 743 voice, 126, 273, 310, 825, 965, 968
acquisition of, 683
Tabassaran, 656–657 actor, 129, 958, 969
Tagalog, 25–26, 28, 30–31, 39, 47–48, 58, 93, 105–106, antipassive, 310, 313–314
117, 275–278, 280, 340, 345, 353, 506, 654 argument modulation, 126, 129, 310
telicity, 221, 305, 344, 419, See also state, change of neutral, 968
template, syntactic, 60–61 passive, 126, 310
combination, 61, 743 PSA modulation, 126, 129
selection principle, 61, 120 undergoer, 129, 958, 971
temporal clauses, 577 VP, 23
tense, 33, 37, 201–202 ellipsis, 24, 498–499
Thai, 270, 494, 496, 529
thematic relations, 9, 107, 242–244 Wari’, 25, 81–82, 190, 198–199
Tibetan, 566 Warlpiri, 83, 273, 280, 324, 327, 334, 682–683
Tibeto-Burman languages, 37, 282–284, 299, 308 Wernicke’s area, 695
Tiwi, 37, 204 wh-questions. See questions
Tlachichilco Tepehua, 309–311, 421 word formation, 383
Toba Batak, 473, 515
Tolmači Karelian, 351 Yagnob, 340–341
tone language, 889 Yaqui, 140–142, 312, 528, 566, 569, 572–574, 580,
Tongan, 193, 296 583, 585
topic, 457, 461–463, 490, 505, 955 Yimas, 861–887
aboutness, 462–463, 465, 469 Yucatec Maya, 68, 296
acceptability scale, 464
contrastive, 24, 471 zero anaphora, 481

Delia Bentley, Ricardo Mairal Usón, Wataru Nakamura, Robert D. V - The Cambridge Handbook of Role and Reference Grammar (2023, Cambridge University Press) - Libgen - Li

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Delia Bentley, Ricardo Mairal Usón, Wataru Nakamura, Robert D. V - The Cambridge Handbook of Role and Reference Grammar (2023, Cambridge University Press) - Libgen - Li

Uploaded by

Copyright:

Available Formats

The Cambridge Handbook of Role and Reference Grammar

Role and Reference Grammar (RRG) is a theory of language in which linguistic

delia bentley is Professor of Romance Linguistics at the University of

ricardo mairal usón is Full Professor of Linguistics at Universidad Nacional

wataru nakamura is Associate Professor of Applied Linguistics at Tohoku

robert d. van valin, jr. is Professor Emeritus in the Departments of Linguistics

Published online by Cambridge University Press

Published online by Cambridge University Press

Published online by Cambridge University Press

Published online by Cambridge University Press

Cambridge University Press is part of Cambridge University Press & Assessment,

Published online by Cambridge University Press

Published online by Cambridge University Press

List of Figures page xi

Introduction Delia Bentley 1

Part One Overview

Part Two Topics in RRG: Simple Sentences

Published online by Cambridge University Press

Part Three Topics in RRG: Complex Sentences

Part Four Applications of RRG

Part Five Grammatical Sketches

Published online by Cambridge University Press

1.1 The monostratal nature of RRG page 19

Published online by Cambridge University Press

1.28 The structure of (65c): English clausal cosubordination

Published online by Cambridge University Press

2.10 Constituent and operator projections in a Japanese

Published online by Cambridge University Press

10.2 Components of the layered structure of the clause (Van

Published online by Cambridge University Press

13.6 Clausal subordination (clause-to-core/complement clause)

Published online by Cambridge University Press

19.2 Simple English transitive linking template 701

Published online by Cambridge University Press

Published online by Cambridge University Press

22.5 Representation of a clause including an argument-

Published online by Cambridge University Press

1.1 Operators page 33

Published online by Cambridge University Press

10.1 Semantic units underlying the syntactic units of the

Published online by Cambridge University Press

26.2 The relationship between the juxtaposed verb in the

Published online by Cambridge University Press

Delia Bentley – University of Manchester

Published online by Cambridge University Press

Semantic representation of simple sentences

Linking syntax and semantics in simple sentences:

Published online by Cambridge University Press

Linking from syntax to semantics: §1.6.2 x x

Structure of complex sentences: §1.3 x (Ch. 13) x (Ch. 13)

Syntax–semantics interface in complex sentences: §1.3.4

Linking syntax and semantics in complex sentences: §1.7

Published online by Cambridge University Press

Aims and Scope of the Cambridge Handbook of Role

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

(Ibáñez Cerda; Toratani), and the encoding of information structure

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

RRG in Modern Linguistic Theory

An important concern of RRG is to engage in cross-theoretical debate. To give

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

Whilst placing due emphasis on relational constructs, RRG has distanced

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

S (actor or undergoer of intransitive), derived-S (of passive)), which, crucially,